WO2021056843A1 - 神经网络训练方法及装置和图像生成方法及装置 - Google Patents

神经网络训练方法及装置和图像生成方法及装置 Download PDF

Info

Publication number
WO2021056843A1
WO2021056843A1 PCT/CN2019/124541 CN2019124541W WO2021056843A1 WO 2021056843 A1 WO2021056843 A1 WO 2021056843A1 CN 2019124541 W CN2019124541 W CN 2019124541W WO 2021056843 A1 WO2021056843 A1 WO 2021056843A1
Authority
WO
WIPO (PCT)
Prior art keywords
distribution
network
discriminant
loss
training
Prior art date
Application number
PCT/CN2019/124541
Other languages
English (en)
French (fr)
Inventor
邓煜彬
戴勃
相里元博
林达华
吕健勤
Original Assignee
北京市商汤科技开发有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京市商汤科技开发有限公司 filed Critical 北京市商汤科技开发有限公司
Priority to SG11202103479VA priority Critical patent/SG11202103479VA/en
Priority to KR1020217010144A priority patent/KR20210055747A/ko
Priority to JP2021518079A priority patent/JP7165818B2/ja
Publication of WO2021056843A1 publication Critical patent/WO2021056843A1/zh
Priority to US17/221,096 priority patent/US20210224607A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G06T9/002—Image coding using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06V10/7747—Organisation of the process, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/776—Validation; Performance evaluation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Definitions

  • the present disclosure relates to the field of computer technology, and in particular to a neural network training method and device, and an image generation method and device.
  • GAN Generative Adversarial Networks
  • Discriminator In related technologies, Generative Adversarial Networks (GAN) is composed of two modules, namely Discriminator and Generator. Inspired by the zero-sum game, the two networks compete with each other to achieve the best generation effect.
  • the discriminator learns to distinguish between real image data and simulated images generated by the generation network by rewarding true targets and penalizing false targets.
  • the generator gradually reduces the punishment of false targets by the discriminator, making the discriminator unable to distinguish real images.
  • the two play games and evolve with each other, and finally achieve the effect of being fake.
  • the generative confrontation network is described by a single scalar output by the discrimination network to describe the authenticity of the input picture, and then the scalar is used to calculate the loss of the network, and then the generative confrontation network is trained.
  • the present disclosure proposes a neural network training method and device, and an image generation method and device.
  • a neural network training method including:
  • the first generated image and the first real image are respectively input to the discriminant network, and the first discriminant distribution of the first generated image and the second discriminant distribution of the first real image are obtained respectively, wherein the first discriminant distribution represents The probability distribution of the real degree of the first generated image, and the second discriminant distribution represents the probability distribution of the real degree of the first real image;
  • the first discriminant distribution the second discriminant distribution, the preset first target distribution, and the preset second target distribution
  • the first network loss of the discriminant network is determined, wherein the first target distribution To generate the target probability distribution of the image, the second target distribution is the target probability distribution of the real image;
  • the generation network and the discriminant network are trained against training.
  • the discriminant network can output the discriminative distribution of the input image, describe the authenticity of the input image in the form of probability distribution, and describe the input image as real from dimensions such as color, texture, proportion, and background.
  • the probability of the image can consider the authenticity of the input image from many aspects, reduce information loss, provide more comprehensive supervision information and more accurate training direction for neural network training, improve training accuracy, and ultimately improve the quality of the generated image, so that the generation
  • the network can be adapted to generate high-definition images.
  • the target probability distribution of the generated image and the target probability distribution of the real image are preset to guide the training process.
  • the real image and the generated image are guided to approach their respective target probability distributions, and the distinction between the real image and the generated image is increased. Enhance the ability of the discriminant network to distinguish between real images and generated images, thereby improving the quality of the images generated by the generating network.
  • the first network loss of the discriminant network is determined according to the first discriminant distribution, the second discriminant distribution, the preset first target distribution, and the preset second target distribution ,include:
  • the target probability distribution of the generated image and the target probability distribution of the real image are preset to guide the training process, and the respective distribution losses are determined respectively, and the real image and the generated image are guided to approach their respective target probabilities during the training process.
  • Distribution increase the distinction between real images and generated images, provide more accurate angle information for the discriminant network, provide more accurate training directions for the discriminant network, and enhance the ability of the discriminant network to distinguish between real images and generated images, thereby improving the generation network The quality of the generated image.
  • determining the first distribution loss of the first generated image according to the first discriminant distribution and the first target distribution includes:
  • mapping the first discriminant distribution to the support set of the first target distribution to obtain a first mapping distribution
  • the first distribution loss is determined.
  • determining the second distribution loss of the first real image according to the second discriminant distribution and the second target distribution includes:
  • the second distribution loss is determined.
  • determining the first network loss according to the first distribution loss and the second distribution loss includes:
  • determining the second network loss of the generating network according to the first discriminant distribution and the second discriminant distribution includes:
  • the second network loss is determined.
  • the generation network can be trained by reducing the difference between the first discriminant distribution and the second discriminant distribution, so that while the performance of the discriminant network is improved, the performance of the generation network is promoted, thereby generating a more realistic generated image , Making the generation network suitable for generating high-definition images.
  • adversarial training of the generation network and the discriminant network includes:
  • the trained generating network and the discriminant network are obtained.
  • adjusting the network parameters of the discrimination network according to the loss of the first network includes:
  • the gradient descent speed of the discriminant network during training can be limited, thereby limiting the training progress of the discriminant network and reducing the probability of the gradient disappearing of the discriminant network , So as to continuously optimize the generation network, improve the performance of the generation network, and make the generated images of the generation network more realistic and suitable for generating high-definition images.
  • adversarial training of the generation network and the discriminant network includes:
  • the first generated image, at least one third generated image, and at least one real image corresponding to the first random vector input to the generating network in the at least one historical training period are respectively input into the discriminant network of the current training period to obtain at least A fourth discriminant distribution of a first generated image, a fifth discriminant distribution of at least one third generated image, and a sixth discriminant distribution of at least one real image;
  • the gradient descent speed of the discriminant network in training can be limited, thereby limiting the training progress of the discriminant network, reducing the probability of the discriminant network appearing gradient disappear, so as to continuously optimize
  • the generation network improves the performance of the generation network, and makes the image generated by the generation network more realistic and suitable for the generation of high-definition images.
  • determining the training progress parameters of the generation network of the current training period according to the fourth discriminant distribution, the fifth discriminant distribution, and the sixth discriminant distribution includes:
  • the ratio of the first difference value to the second difference value is determined as the training progress parameter of the generating network of the current training period.
  • an image generation method including:
  • the third random vector is input into the generating network obtained after training for processing, and the target image is obtained.
  • a neural network training device including:
  • a generating module which is used to input the first random vector into the generating network to obtain the first generated image
  • the discrimination module is used to input the first generated image and the first real image into a discrimination network respectively to obtain the first discriminant distribution of the first generated image and the second discriminant distribution of the first real image, wherein the The first discriminant distribution represents the probability distribution of the real degree of the first generated image, and the second discriminant distribution represents the probability distribution of the real degree of the first real image;
  • the first determining module is configured to determine the first network loss of the discriminant network according to the first discriminant distribution, the second discriminant distribution, the preset first target distribution, and the preset second target distribution, where ,
  • the first target distribution is the target probability distribution of the generated image, and the second target distribution is the target probability distribution of the real image;
  • a second determining module configured to determine the second network loss of the generating network according to the first discriminant distribution and the second discriminant distribution
  • the training module is used to counter-train the generation network and the discrimination network according to the loss of the first network and the loss of the second network.
  • the first determining module is further configured to:
  • the first determining module is further configured to:
  • mapping the first discriminant distribution to the support set of the first target distribution to obtain a first mapping distribution
  • the first distribution loss is determined.
  • the first determining module is further configured to:
  • the second distribution loss is determined.
  • the first determining module is further configured to:
  • the second determining module is further configured to:
  • the second network loss is determined.
  • the training module is further configured to:
  • the trained generating network and the discriminant network are obtained.
  • the training module is further configured to:
  • the training module is further configured to:
  • the first generated image, at least one third generated image, and at least one real image corresponding to the first random vector input to the generating network in the at least one historical training period are respectively input into the discriminant network of the current training period to obtain at least A fourth discriminant distribution of a first generated image, a fifth discriminant distribution of at least one third generated image, and a sixth discriminant distribution of at least one real image;
  • the training module is further configured to:
  • the ratio of the first difference value to the second difference value is determined as the training progress parameter of the generating network of the current training period.
  • an image generation device which includes:
  • the obtaining module is configured to input the third random vector into the generating network obtained after training for processing to obtain a target image.
  • an electronic device including:
  • a memory for storing processor executable instructions
  • the processor is configured to execute the above method.
  • a computer-readable storage medium having computer program instructions stored thereon, and the computer program instructions implement the above method when executed by a processor.
  • a computer program including computer readable code, and when the computer readable code is executed in an electronic device, a processor in the electronic device is executed to execute the above-mentioned method.
  • Fig. 1 shows a flowchart of a neural network training method according to an embodiment of the present disclosure
  • Fig. 2 shows a schematic diagram of the application of a neural network training method according to an embodiment of the present disclosure
  • Fig. 3 shows a block diagram of a neural network training device according to an embodiment of the present disclosure
  • Fig. 4 shows a block diagram of an electronic device according to an embodiment of the present disclosure
  • FIG. 5 shows a block diagram of an electronic device according to an embodiment of the present disclosure.
  • Fig. 1 shows a flowchart of a neural network training method according to an embodiment of the present disclosure. As shown in Fig. 1, the method includes:
  • step S11 the first random vector is input to the generating network to obtain the first generated image
  • step S12 the first generated image and the first real image are respectively input to a discriminant network, and the first discriminant distribution of the first generated image and the second discriminant distribution of the first real image are obtained respectively, wherein the The first discriminant distribution represents the probability distribution of the real degree of the first generated image, and the second discriminant distribution represents the probability distribution of the real degree of the first real image;
  • step S13 the first network loss of the discriminant network is determined according to the first discriminant distribution, the second discriminant distribution, the preset first target distribution, and the preset second target distribution, wherein The first target distribution is the target probability distribution of the generated image, and the second target distribution is the target probability distribution of the real image;
  • step S14 determine the second network loss of the generating network according to the first discriminant distribution and the second discriminant distribution;
  • step S15 according to the first network loss and the second network loss, the generation network and the discriminant network are trained against training.
  • the discriminant network can output the discriminative distribution of the input image, describe the authenticity of the input image in the form of probability distribution, and describe the input image as real from dimensions such as color, texture, proportion, and background.
  • the probability of the image can consider the authenticity of the input image from many aspects, reduce information loss, provide more comprehensive supervision information and more accurate training direction for neural network training, improve training accuracy, and ultimately improve the quality of the generated image, so that the generation
  • the network can be adapted to generate high-definition images.
  • the target probability distribution of the generated image and the target probability distribution of the real image are preset to guide the training process.
  • the real image and the generated image are guided to approach their respective target probability distributions, and the distinction between the real image and the generated image is increased. Enhance the ability of the discriminant network to distinguish between real images and generated images, thereby improving the quality of the images generated by the generating network.
  • the neural network training method may be executed by a terminal device or other processing equipment, where the terminal device may be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone , Cordless phones, Personal Digital Assistant (PDA), handheld devices, computing devices, vehicle-mounted devices, wearable devices, etc.
  • UE user equipment
  • PDA Personal Digital Assistant
  • Other processing devices can be servers or cloud servers.
  • the neural network training method can be implemented by a processor calling computer-readable instructions stored in a memory.
  • the neural network may be a generative confrontation network composed of a generative network and a discriminant network.
  • the generation network may be a deep learning neural network such as a convolutional neural network, and the present disclosure does not limit the type and structure of the generation network.
  • the discriminant network may be a deep learning neural network such as a convolutional neural network, and the present disclosure does not limit the type and structure of the discriminant network.
  • the generation network can process the random vector to obtain the generated image.
  • the random vector can be a vector with random numbers for each element, and can be obtained by random sampling and other methods.
  • the first random vector may be obtained by random sampling or the like, and the generation network may perform processing such as convolution on the first random vector to obtain a first generated image corresponding to the first random vector.
  • the first random vector is a randomly generated vector, therefore, the first generated image is a random image.
  • the first real image may be any real image, for example, it may be a real image captured by an image acquisition device (for example, a camera, a camera, etc.).
  • the first real image and the first generated image can be input into the discriminant network respectively, and the first discriminant distribution of the first generated image and the second discriminant distribution of the first real image are obtained respectively.
  • the first discriminant distribution and the second discriminant distribution are obtained.
  • the discriminant distribution can be a parameter in the form of a vector, for example, a probability distribution can be expressed in the form of a vector.
  • the first discriminant distribution may indicate the degree of authenticity of the first generated image, that is, the probability that the first generated image is a real image may be described by the first discriminant distribution.
  • the second discriminant distribution may indicate the degree of reality of the first real image, that is, the probability that the first real image is a real image can be described by the second discriminant distribution.
  • the authenticity of the image is described in the form of a distribution (such as a multi-dimensional vector).
  • the authenticity of the image can be considered from multiple aspects such as color, texture, proportion, background, etc., to reduce information loss and provide accurate training directions for training.
  • the target probability distribution of the real image (ie, the second target distribution) and the target probability distribution of the generated image (ie, the first target distribution) can be preset.
  • the network loss corresponding to the generated image and the network loss corresponding to the real image can be determined according to the target probability distribution of the real image and the target probability distribution of the generated image, and the network loss corresponding to the generated image and the network loss corresponding to the real image can be used respectively Adjust the parameters of the discriminant network so that the second discriminant distribution of the real image is close to the second target distribution and is significantly different from the first target distribution, and the first discriminant distribution of the generated image is close to the first target distribution and is similar to the second target distribution.
  • There is a significant difference in distribution which can increase the degree of discrimination between real images and generated images, enhance the ability of the discrimination network to distinguish between real images and generated images, and thereby improve the quality of the images generated by the generation network.
  • the anchor distribution of the generated image (ie, the first target distribution) and the anchor distribution of the real image (ie, the second target distribution) can be preset, and the vector representing the anchor distribution of the generated image and the real image
  • the vector of the anchor distribution has a significant difference.
  • the network parameters of the discriminant network can be adjusted to reduce the difference between the first discriminant distribution and the anchor distribution of the generated image. In this process, the difference between the first discriminant distribution and the anchor distribution of the real image will increase.
  • the difference between the second discriminant distribution and the anchor distribution of the real image is also reduced.
  • the difference between the second discriminant distribution and the anchor distribution of the generated image will increase. That is, the anchor distributions are respectively preset for the real image and the generated image, so that the distribution difference between the real image and the generated image is increased, so as to improve the distinguishing ability of the discrimination network between the real image and the generated image.
  • step S13 may include: determining the first distribution loss of the first generated image according to the first discriminant distribution and the first target distribution; according to the second discriminant distribution and The second target distribution determines the second distribution loss of the first real image; the first network loss is determined according to the first distribution loss and the second distribution loss.
  • the first target distribution is an accurate probability distribution, and the difference between the first target distribution and the first discriminant distribution can be determined, so as to determine the first distribution loss.
  • the network loss corresponding to the first generated image may be determined according to the first discriminant distribution and the first target distribution.
  • determining the first distribution loss of the first generated image according to the first discriminant distribution and the first target distribution includes: mapping the first discriminant distribution to a support set of the first target distribution , Obtain a first mapping distribution; determine a first relative entropy of the first mapping distribution and the first target distribution; determine the first distribution loss according to the first relative entropy.
  • the support set of the first discriminant distribution and the first target distribution may be different, that is, the distribution range of the first discriminant distribution is different from that of the first target distribution.
  • the distribution range of the first target distribution is different.
  • the first discriminant distribution can be mapped to the support set of the first target distribution, or the first target distribution can be mapped to the support set of the first discriminant distribution.
  • the first discriminant distribution can be projected by means of linear transformation, for example, the projection matrix can be used to map the first discriminant distribution to the support set of the first target distribution, that is, the vector of the first discriminant distribution can be Linear transformation, the vector obtained after transformation is the first mapping distribution after mapping to the support set of the first target distribution.
  • the first relative entropy between the first mapping distribution and the first target distribution may be determined, and the first relative entropy may represent two items in the same support set.
  • the difference in probability distribution ie, the difference between the first mapping distribution and the first target distribution.
  • the difference between the first mapping distribution and the first target distribution may also be determined by other methods such as JS divergence (Jensen-Shannon divergence) or Wasserstein distance.
  • the first distribution loss (that is, the network loss corresponding to the generated image) may be determined according to the first relative entropy.
  • the first relative entropy may be determined as the first distribution loss, or the first relative entropy may be subjected to arithmetic processing, for example, the first relative entropy may be weighted, logarithmic, exponential, etc., to obtain the first relative entropy.
  • the first distribution loss does not limit the method for determining the first distribution loss.
  • the second target distribution is an accurate probability distribution, and the difference between the second target distribution and the second discriminant distribution can be determined, so as to determine the second distribution loss.
  • the network loss corresponding to the first real image can be determined according to the second discriminant distribution and the second target distribution.
  • determining the second distribution loss of the first real image according to the second discriminant distribution and the second target distribution includes: mapping the second discriminant distribution to a support set of the second target distribution , Obtain a second mapping distribution; determine a second relative entropy of the second mapping distribution and the second target distribution; determine the second distribution loss according to the second relative entropy.
  • the support set of the second discriminant distribution and the second target distribution may be different, that is, the distribution range of the second discriminant distribution is different from that of the second target distribution.
  • the distribution range of the second target distribution is different.
  • the second discriminant distribution can be mapped to the support set of the second target distribution, or the second target distribution can be mapped to the support set of the second discriminant distribution, or the second discriminant distribution and the second target distribution can be mapped to the same support set , So that the distribution range of the second discriminant distribution is the same as the distribution range of the second target distribution, and the difference between the two probability distributions can be compared in the same distribution range.
  • the second discriminant distribution can be projected by means of linear transformation, for example, the projection matrix can be used to map the second discriminant distribution to the support set of the second target distribution, that is, the vector of the second discriminant distribution can be performed Linear transformation, the vector obtained after transformation is the second mapping distribution after mapping to the support set of the second target distribution.
  • the second relative entropy of the second mapping distribution and the second target distribution may be determined, and the second relative entropy may represent the difference between the two probability distributions in the same support set (ie, the second mapping The difference between the distribution and the second target distribution).
  • the calculation method of the second relative entropy is similar to the first relative entropy, and will not be repeated here.
  • the difference between the second mapping distribution and the second target distribution can also be determined by other methods such as JS divergence (Jensen-Shannon divergence) or Wasserstein distance.
  • the second distribution loss (that is, the network loss corresponding to the generated image) may be determined according to the second relative entropy.
  • the second relative entropy may be determined as the second distribution loss, or the second relative entropy may be subjected to arithmetic processing, for example, the second relative entropy may be weighted, logarithmic, exponential, etc., to obtain the The second distribution loss.
  • the present disclosure does not limit the determination method of the second distribution loss.
  • the first network loss may be determined according to the first distribution loss of the first generated image and the second distribution loss of the second generated image.
  • determining the first network loss according to the first distribution loss and the second distribution loss includes: performing a weighted sum process on the first distribution loss and the second distribution loss to obtain the The first network loss.
  • the weights of the first distribution loss and the second distribution loss can be the same, that is, the first network loss can be obtained by directly summing the first distribution loss and the second distribution loss.
  • the weights of the first distribution loss and the second distribution loss may be different, that is, the first distribution loss and the second distribution loss are respectively multiplied by their respective weights and then summed to obtain the first network loss.
  • the weights of the first distribution loss and the second distribution loss may be preset, and the present disclosure does not limit the weights of the first distribution loss and the second distribution loss.
  • the target probability distribution of the generated image and the target probability distribution of the real image are preset to guide the training process, and the respective distribution losses are determined respectively, and the real image and the generated image are guided to approach their respective target probabilities during the training process.
  • Distribution increase the distinction between real images and generated images, provide more accurate angle information for the discriminant network, provide more accurate training directions for the discriminant network, and enhance the ability of the discriminant network to distinguish between real images and generated images, thereby improving the generation network The quality of the generated image.
  • the second network loss of the generating network can also be determined.
  • the discriminant network needs to discriminate whether the input image is a real image or a generated image. Therefore, the discriminant network can enhance the ability to distinguish between the real image and the generated image during the training process, that is, make the discrimination distribution of the real image and the generated image close to each other. The probability distribution of the target, thereby increasing the degree of discrimination between the real image and the generated image.
  • the goal of the generation network is to make the generated image close to the real image, that is, to make the generated image realistic enough to make it difficult for the discrimination network to distinguish the generated image output by the generation network.
  • the performance of the discriminant network and the generation network are strong, that is, the discriminant network has a strong discriminative ability and can distinguish between real images and low-fidelity generated images, and the images generated by the generation network are realistic The degree is so high that it is difficult for the discriminant network to distinguish high-quality generated images.
  • the performance improvement of the discrimination network can promote the performance improvement of the generation network, that is, the stronger the ability of the discrimination network to distinguish the real image and the generated image, the higher the fidelity of the image generated by the generation network.
  • step S14 may include: determining a third relative entropy of the first discriminant distribution and the second discriminant distribution; and determining the second network loss according to the third relative entropy.
  • the third relative entropy of the first discriminant distribution and the second discriminant distribution can be determined, and the third relative entropy represents the difference between the two probability distributions in the same support set (ie, the third mapping distribution The difference with the fourth mapping distribution).
  • the calculation method of the third relative entropy is similar to that of the first relative entropy, and will not be repeated here.
  • the difference between the first discriminant distribution and the second discriminant distribution can also be determined by other methods such as JS divergence (Jensen-Shannon divergence) or Wasserstein distance, so as to determine the network loss of the generated network through the difference between the two .
  • the second network loss may be determined according to the third relative entropy.
  • the third relative entropy can be determined as the second network loss, or the third relative entropy can be calculated, for example, the third relative entropy can be weighted, logarithm, exponent, etc., to obtain the second network loss.
  • the present disclosure does not limit the method for determining the loss of the second network.
  • the support sets of the first discriminant distribution and the second discriminant distribution are different, that is, the distribution ranges of the first discriminant distribution and the second discriminant distribution may be different.
  • the support set of the first discriminant distribution and the second discriminant distribution can be overlapped by linear transformation.
  • the first discriminant distribution and the second discriminant distribution can be mapped to the target support set, so that the distribution range of the second discriminant distribution is the same as that of the first discriminant distribution.
  • the distribution range of the distribution is the same, and the difference between the two probability distributions can be compared in the same distribution range.
  • the target support set is the support set of the first discriminant distribution or the support set of the second discriminant distribution.
  • the second discriminant distribution can be mapped to the support set of the first discriminant distribution by means of linear transformation, that is, the vector of the second discriminant distribution can be linearly transformed, and the vector obtained after the transformation is the one mapped to the first discriminant distribution Support the fourth mapping distribution after the set, and use the first discriminant distribution as the third mapping distribution.
  • the first discriminant distribution can be mapped to the support set of the second discriminant distribution by means of linear transformation, that is, the vector of the first discriminant distribution can be linearly transformed, and the vector obtained after the transformation is mapped to the second discriminant
  • the third mapping distribution after the distributed support set, and the second discriminant distribution is used as the fourth mapping distribution.
  • the target support set can also be other support sets.
  • a support set can be preset, and both the first discriminant distribution and the second discriminant distribution can be mapped to the support set to obtain the third mapping distribution and The fourth mapping distribution. Further, the third relative entropy of the third mapping distribution and the fourth mapping distribution can be calculated.
  • the present disclosure does not limit the target support set.
  • the generation network can be trained by reducing the difference between the first discriminant distribution and the second discriminant distribution, so that while the performance of the discriminant network is improved, the performance of the generation network is promoted, thereby generating a more realistic generated image , Making the generation network suitable for generating high-definition images.
  • the training generation network and the discrimination network can be opposed to the training generation network and the discrimination network according to the first network loss of the discrimination network and the second network loss of the generation network. That is, through training, the performance of the generation network and the discrimination network are simultaneously improved, the discrimination ability of the discrimination network is improved, and the ability of the generation network to generate images with higher fidelity is improved, and the generation network and the discrimination network are in a balanced state.
  • step S15 may include: adjusting the network parameters of the discrimination network according to the first network loss; adjusting the network parameters of the generating network according to the second network loss; In the case that the generating network satisfies the training condition, the trained generating network and the discriminating network are obtained.
  • the training progress of the discriminant network is usually ahead of the generation network. If the discriminant network progress is faster and the training is completed in advance, the generation network cannot be provided with the gradient in the back propagation. Therefore, the parameters of the generated network cannot be updated, that is, the performance of the generated network cannot be improved. Therefore, the performance of the image generated by the generation network is limited, it is not suitable for generating high-definition images, and the fidelity is low.
  • adjusting the network parameters of the discrimination network according to the first network loss includes: inputting a second random vector into a generating network to obtain a second generated image; and interpolating a second real image according to the second generated image Processing to obtain an interpolated image; input the interpolated image into the discriminant network to obtain a third discriminant distribution of the interpolated image; determine the gradient of the network parameter of the discriminant network according to the third discriminant distribution; In the case where the gradient is greater than the gradient threshold, the gradient penalty parameter is determined according to the third discriminant distribution; and the network parameter of the discriminant network is adjusted according to the first network loss and the gradient penalty parameter.
  • the second random vector may be obtained through random sampling or the like, and input into the generating network to obtain the second generated image, that is, to obtain an unreal image.
  • the second generated image can also be obtained in other ways, for example, a non-real image can be directly generated randomly.
  • the second generated image and the second real image may be subjected to interpolation processing to obtain an interpolated image, that is, the interpolated image is a composite image of the real image and the non-real image, and the interpolated image includes some Real images, including some non-real images.
  • random nonlinear interpolation may be performed on the second real image and the second generated image to obtain the interpolated image.
  • the present disclosure does not limit the method of obtaining the interpolated image.
  • the interpolated image can be input to the discriminant network to obtain the third discriminant distribution of the interpolated image, that is, the discriminant network can perform discrimination processing on the composite image of the real image and the unreal image to obtain the third discriminant distributed.
  • the third discriminant distribution can be used to determine the gradient of the network parameters of the discriminant network.
  • the target probability distribution of the interpolated image can be preset (for example, the probability that the interpolated image is a real image is 50 % Target probability distribution), and use the third discriminant distribution and the relative entropy of the target probability distribution to determine the gradient of the discriminant network's network parameters.
  • the relative entropy of the third discriminant distribution and the target probability distribution can be backpropagated, and the relative entropy and the partial differential of each network parameter of the discriminant network can be calculated to obtain the gradient of the network parameter.
  • other types of differences such as the JS divergence of the third discriminant distribution and the target probability distribution can also be used to determine the parameter gradient of the discriminant network.
  • the gradient penalty parameter can be determined according to the third discriminant distribution.
  • the gradient threshold can be a threshold that limits the gradient. If the gradient is large, the gradient may fall faster during the training process (that is, the training step is larger, and the network loss tends to the minimum speed faster), Therefore, the gradient can be restricted by the gradient threshold.
  • the gradient threshold may be set to 10, 20, etc., and the present disclosure does not limit the gradient threshold.
  • the gradient penalty parameter is used to adjust the gradient of the network parameter that exceeds the gradient threshold, or limit the gradient descent speed, so that the gradient of the parameter is smoother and the gradient descent speed is slowed down.
  • the gradient penalty parameter can be determined according to the expected value of the third discriminant distribution.
  • the gradient penalty parameter can be a compensation parameter for gradient descent.
  • the gradient penalty parameter can be used to adjust the partial differential multiplier, or the gradient penalty parameter can be used to change the direction of gradient descent to limit the gradient, thereby reducing the network of the discriminant network
  • the gradient descent speed of the parameter prevents the gradient of the discriminant network from dropping too fast, causing the discriminant network to converge prematurely (that is, the training is completed too quickly).
  • the third discriminant distribution is a probability distribution
  • the expected value of the probability distribution can be calculated
  • the gradient penalty parameter can be determined according to the expected value.
  • the expected value can be determined as the multiplier of the partial differential of the network parameter, that is, The expected value is determined as the gradient penalty parameter, and the gradient penalty parameter is used as the gradient multiplier.
  • the present disclosure does not limit the determination method of the gradient penalty parameter.
  • the network parameters of the discrimination network can be adjusted according to the first network loss and gradient penalty parameters. That is, in the process of backpropagating the loss of the first network to make the gradient drop, the gradient penalty parameter is added to adjust the network parameters of the discriminant network while preventing the gradient from dropping too fast, that is, preventing the discriminant network from being trained prematurely.
  • the gradient penalty parameter can be used as the multiplier of the partial differential, that is, the multiplier of the gradient, so as to slow down the gradient descent speed and prevent the judgment that the network is trained too early.
  • the network parameter of the judgment network can be adjusted according to the loss of the first network, that is, the loss of the first network is back-propagated. Gradient descent reduces the loss of the first network.
  • the network parameters of the discriminant network when adjusting the network parameters of the discriminant network, check whether the gradient of the discriminant network is greater than or equal to the gradient threshold, and set the gradient penalty parameter when the gradient of the discriminant network is greater than or equal to the gradient threshold. It is also possible not to check the gradient of the discriminant network, but to control the training progress of the discriminant network in other ways (for example, suspend the adjustment of the network parameters of the discriminant network, and only adjust the network parameters of the generated network, etc.).
  • the gradient descent speed of the discriminant network during training can be limited, thereby limiting the training progress of the discriminant network and reducing the probability of the gradient disappearing of the discriminant network , So as to continuously optimize the generation network, improve the performance of the generation network, and make the generated images of the generation network more realistic and suitable for generating high-definition images.
  • the network parameters of the generation network can be adjusted according to the second network loss.
  • the loss of the second network is back-propagated to decrease the gradient, so that the loss of the second network is reduced, so as to improve the performance of the generation network. performance.
  • the training discriminant network and the generation network can fight against the training discriminant network and the generation network.
  • the network parameters of the discrimination network are adjusted through the loss of the first network, the network parameters of the generation network remain unchanged, and the generation network is adjusted by the second network loss.
  • the network parameters of the network the network parameters of the judgment network remain unchanged.
  • the above training process can be performed iteratively until the discrimination network and the generation network meet the training conditions.
  • the training conditions include the discrimination network and the generation network reaching a balanced state.
  • the network loss of the discrimination network and the generation network is less than or equal to the expected Set a threshold, or converge to a preset interval.
  • the training conditions include the following two conditions to achieve a balanced state: first, the network loss of the generating network is less than or equal to a preset threshold or converges to a preset interval; second, the input image represented by the discriminant distribution of the discriminant network output is The probability of the real image is maximized. At this time, the ability to distinguish between real images and generated images by the network is strong, and the images generated by the generation network are of higher quality and fidelity.
  • the training progress of the discriminant network can also be controlled to reduce the probability of gradient disappearance of the discriminant network.
  • step S15 may include: inputting the first random vector input to the generating network in at least one historical training period into the generating network of the current training period to obtain at least one third generated image; The first generated image corresponding to the first random vector of the generating network, the at least one third generated image, and the at least one real image are respectively input into the discriminant network of the current training period, and the fourth discriminant distribution of the at least one first generated image, The fifth discriminant distribution of at least one third generated image and the sixth discriminant distribution of at least one real image; the generation network of the current training period is determined according to the fourth discriminant distribution, the fifth discriminant distribution, and the sixth discriminant distribution If the training progress parameter is less than or equal to the training progress threshold, stop adjusting the network parameters of the discrimination network, and only adjust the network parameters of the generating network.
  • a buffer area can be opened during the training process, for example, an experience buffer, in which at least one (for example, M, M is a positive integer) can be stored
  • the first random vector of the historical training period and the M first generated images generated by the generating network according to the first random vector in the above-mentioned M historical training periods, that is, each historical training period can generate a first random vector through a first random vector.
  • the first random vector of M historical training periods and the generated M first generated images can be stored in the buffer area.
  • the first random vector and first generated image of the latest training cycle can be used to replace the first random vector and first generated image stored in the buffer area earliest.
  • the first random vector input to the generating network in at least one historical training period may be input to the generating network of the current training period to obtain at least one third generated image.
  • the m (m is less than or equal to M, and m is a positive integer) first random vectors are input to the generating network of the current training period, and m third generated images are obtained.
  • the m third generated images may be separately subjected to the discrimination processing through the discriminant network of the current training period to obtain m fifth discriminant distributions.
  • the first generated images of m historical training periods can be discriminated respectively through the discriminant network of the current training period to obtain m fourth discriminant distributions.
  • M real images can be randomly sampled from the database, and the m real images can be discriminated respectively through the discriminant network of the current training period to obtain m sixth discriminant distributions.
  • the training progress parameters of the generation network of the current training period can be determined according to m fourth discriminant distributions, m fifth discriminant distributions, and m sixth discriminant distributions, that is, determine the discriminant network Whether the training progress is significantly ahead of the generative network, and if a significant lead is determined, adjust the training progress parameters of the generative network to improve the training progress of the generative network and reduce the difference in the training progress of the discriminant network and the generative network, that is, pause the discriminant network
  • the generation network is trained separately to increase the progress parameters of the generation network and speed up the progress.
  • determining the training progress parameter of the generation network of the current training period according to the fourth discriminant distribution, the fifth discriminant distribution, and the sixth discriminant distribution includes: obtaining at least one of the The first expected value of the fourth discriminant distribution, the second expected value of at least one of the fifth discriminant distribution, and the third expected value of at least one of the sixth discriminant distribution; the first average of the at least one first expected value is obtained respectively Value, at least one second average value of the second expected value, and at least one third average value of the third expected value; determine the first difference between the third average value and the second average value, and the The second difference between the second average value and the first average value; and the ratio of the first difference value and the second difference value is determined as the training progress parameter of the generating network of the current training period.
  • the expected values of m fourth discriminant distributions can be calculated respectively to obtain m first expected values
  • the expected values of m fifth discriminant distributions can be calculated respectively
  • m second expected values can be obtained
  • Calculate the expected values of m sixth discriminant distributions respectively and obtain m third expected values.
  • the m first expected values can be averaged to obtain the first average value S B
  • the m second expected values can be averaged to obtain the second average value S G
  • the m-th expected value can be averaged.
  • the three expected values are averaged to obtain the third average value S R.
  • the first difference between the third average value and the second average value can be determined, and the second difference value between the second average value and the first average value ( S G -S B ).
  • the ratio of the first difference value may be a second difference (S R -S G) / ( S G -S B) is determined as the parameters of the current training schedule generation network training period.
  • the preset number of training times can also be used as the training progress parameter of the generation network. For example, the generation network and the discriminant network can be trained together for 100 times, the discriminant network training can be suspended, and the generation network can be trained separately for 50 times. Then make the generation network and the discriminant network train 100 times together... until the generation network and the discriminant network meet the training conditions.
  • a training progress threshold can be set.
  • the training progress threshold is a threshold for determining the training progress of the generated network. If the training progress parameter is less than or equal to the training progress threshold, it indicates that the training progress of the discriminating network is significantly ahead
  • the adjustment of the network parameters of the discrimination network can be suspended, and only the network parameters of the generation network can be adjusted.
  • the network parameters of the discrimination network and the generation network can be adjusted at the same time, that is, Pause the training of the discriminant network for at least one training cycle, and only train the generation network (that is, adjust the network parameters of the generation network only according to the third network loss, and keep the network parameters of the discrimination network unchanged), until the training progress of the generation network is close to the discrimination network The training progress, and then confront the training generation network and the discriminant network.
  • the training speed of the discriminant network can also be reduced, such as extending the training period of the discriminant network or reducing the gradient descent speed of the discriminant network, etc., until the training progress parameter If it is greater than the training progress threshold, the training speed of the discriminant network can be restored.
  • the gradient descent speed of the discriminant network in training can be limited, thereby limiting the training progress of the discriminant network, reducing the probability of the discriminant network appearing gradient disappear, so as to continuously optimize
  • the generation network improves the performance of the generation network, and makes the image generated by the generation network more realistic and suitable for the generation of high-definition images.
  • the generating network can be used to generate an image, and the generated image has a higher fidelity.
  • the present disclosure also provides an image generation method that uses the generated confrontation network completed by the above training to generate an image.
  • an image generation method includes: obtaining a third random vector; and inputting the third random vector into the generation network obtained after training of the neural network training method described above for processing to obtain a target image.
  • the third random vector can be obtained by random sampling, etc., and input the third random vector into the trained generation network.
  • the generation network can output target images with high fidelity.
  • the target image may be a high-definition image, that is, the trained generation network may be suitable for generating a high-definition image with high fidelity.
  • the discriminant network can discriminate the distribution of the input image output, describe the authenticity of the input image in the form of distribution, consider the authenticity of the input image from multiple aspects, reduce information loss, and be neural Network training provides more comprehensive supervision information and more accurate training directions, improves training accuracy, and improves the quality of generated images, making the generation network suitable for generating high-definition images.
  • the target probability distribution of the generated image and the target probability distribution of the real image are preset to guide the training process, and their respective distribution losses are determined respectively.
  • the real image and the generated image are guided to approach their respective target probability distributions, increasing
  • the distinction between real images and generated images enhances the ability of the discriminant network to distinguish between real images and generated images, and trains the generation network by reducing the difference between the first discriminant distribution and the second discriminant distribution, so that while the performance of the discriminant network is improved, Promote the performance improvement of the generation network, thereby generating a higher degree of fidelity generated images, making the generation network suitable for generating high-definition images.
  • the gradient descent speed of the discriminant network during training by checking whether the gradient of the discriminant network's network parameters is greater than or equal to the gradient threshold, or checking the training progress of the discriminant network and the generating network, thereby limiting the training of the discriminant network Progress, reduce the probability of the disappearance of the gradient of the discrimination network, thereby continuously optimizing the generation network, improving the performance of the generation network, making the generated image of the generation network more realistic, and suitable for generating high-definition images.
  • Fig. 2 shows an application schematic diagram of a neural network training method according to an embodiment of the present disclosure.
  • a first random vector can be input to a generating network, and the generating network can output a first generated image.
  • the discriminant network may perform discrimination processing on the first generated image and the first real image respectively, and obtain the first discriminant distribution of the first generated image and the second discriminant distribution of the first real image respectively.
  • the anchor distribution of the generated image (that is, the first target distribution) and the anchor distribution of the real image (that is, the second target distribution) can be preset.
  • the first distribution loss corresponding to the first generated image can be determined according to the first discriminant distribution and the first target distribution.
  • the second distribution loss corresponding to the first real image can be determined.
  • the first network loss of the discrimination network can be determined by the first distribution loss and the second distribution loss.
  • the second network loss of the generating network can be determined by the first discriminant distribution and the second discriminant distribution. Further, the first network loss and the second network loss can be used to fight the training generation network and the discriminant network. That is, the network parameters of the judgment network are adjusted through the first network loss, and the network parameters of the generated network are adjusted through the second network loss.
  • the training progress of the discriminant network is usually faster than that of the generation network.
  • the generation network cannot continue to be optimized.
  • the training progress of the discriminant network can be controlled by detecting the gradient of the discriminant network.
  • a real image and the generated image can be interpolated, and the third discriminant distribution of the interpolated image can be determined by the discriminant network, and then according to the first The expected value of the three discriminant distribution determines the gradient penalty parameter.
  • the gradient of the discriminant network is greater than or equal to the preset gradient threshold, in order to prevent the gradient of the discriminant network from falling too fast, causing the discriminant network to be trained too quickly, you can perform the loss of the first network In the process of backpropagation making the gradient descent, a gradient penalty parameter is added to limit the gradient descent speed of the discriminant network.
  • the training progress of the discriminant network and the generating network can also be checked.
  • the M first random vectors input to the generating network in M historical training periods can be input into the generating network of the current training period to obtain M third generated images.
  • the training progress parameters of the generation network of the current training period are determined. If the training progress parameter is less than or equal to the training progress threshold, it indicates that the training progress of the discriminant network is significantly ahead of the generation network, and the adjustment of the network parameters of the discriminant network can be suspended, and only the network parameters of the generation network can be adjusted.
  • the generation network may be used to generate the target image, and the target image may be a high-definition image with relatively high fidelity.
  • the neural network training method can enhance the stability of the generated confrontation and the quality and fidelity of the generated image. It can be applied to scenes such as the generation or synthesis of scenes in games, the transfer or conversion of image styles, and image clustering.
  • the present disclosure does not limit the usage scenarios of the neural network training method.
  • Fig. 3 shows a block diagram of a neural network training device according to an embodiment of the present disclosure. As shown in Fig. 3, the device includes:
  • the generating module 11 is configured to input the first random vector into the generating network to obtain the first generated image
  • the discrimination module 12 is configured to input the first generated image and the first real image into a discrimination network respectively, and obtain the first discriminant distribution of the first generated image and the second discriminant distribution of the first real image, respectively.
  • the first discriminant distribution represents the probability distribution of the real degree of the first generated image
  • the second discriminant distribution represents the probability distribution of the real degree of the first real image
  • the first determining module 13 is configured to determine the first network loss of the discriminant network according to the first discriminant distribution, the second discriminant distribution, the preset first target distribution, and the preset second target distribution, wherein, the first target distribution is the target probability distribution of the generated image, and the second target distribution is the target probability distribution of the real image;
  • the second determining module 14 is configured to determine the second network loss of the generating network according to the first discriminant distribution and the second discriminant distribution;
  • the training module 15 is used to counter-train the generation network and the discrimination network according to the loss of the first network and the loss of the second network.
  • the first determining module is further configured to:
  • the first determining module is further configured to:
  • mapping the first discriminant distribution to the support set of the first target distribution to obtain a first mapping distribution
  • the first distribution loss is determined.
  • the first determining module is further configured to:
  • the second distribution loss is determined.
  • the first determining module is further configured to:
  • the second determining module is further configured to:
  • the second network loss is determined.
  • the training module is further configured to:
  • the trained generating network and the discriminant network are obtained.
  • the training module is further configured to:
  • the training module is further configured to:
  • the first generated image, at least one third generated image, and at least one real image corresponding to the first random vector input to the generating network in the at least one historical training period are respectively input into the discriminant network of the current training period to obtain at least A fourth discriminant distribution of a first generated image, a fifth discriminant distribution of at least one third generated image, and a sixth discriminant distribution of at least one real image;
  • the training module is further configured to:
  • the ratio of the first difference value to the second difference value is determined as the training progress parameter of the generating network of the current training period.
  • the present disclosure also provides an image generation device that uses the generated confrontation network completed by the above training to generate images.
  • an image generation device includes:
  • the obtaining module is configured to input the third random vector into the generating network obtained after training for processing to obtain a target image.
  • the present disclosure also provides neural network training devices, electronic equipment, computer-readable storage media, and programs, all of which can be used to implement any neural network training method provided in the present disclosure.
  • the writing order of the steps does not mean a strict execution order but constitutes any limitation on the implementation process.
  • the specific execution order of each step should be based on its function and possibility.
  • the inner logic is determined.
  • the functions or modules contained in the device provided in the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments.
  • specific implementation refer to the description of the above method embodiments. For brevity, here No longer.
  • the embodiments of the present disclosure also provide a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the above-mentioned method when executed by a processor.
  • the computer-readable storage medium may be a volatile computer-readable storage medium or a non-volatile computer-readable storage medium.
  • An embodiment of the present disclosure also provides an electronic device, including: a processor; a memory for storing executable instructions of the processor; wherein the processor is configured as the above method.
  • the electronic device can be provided as a terminal, server or other form of device.
  • Fig. 4 is a block diagram showing an electronic device 800 according to an exemplary embodiment.
  • the electronic device 800 may be a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, and other terminals.
  • the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power component 806, a multimedia component 808, an audio component 810, an input/output (I/O) interface 812, and a sensor component 814 , And communication component 816.
  • the processing component 802 generally controls the overall operations of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations.
  • the processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the foregoing method.
  • the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components.
  • the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.
  • the memory 804 is configured to store various types of data to support operations in the electronic device 800. Examples of these data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc.
  • the memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable and Programmable read only memory (EPROM), programmable read only memory (PROM), read only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.
  • SRAM static random access memory
  • EEPROM electrically erasable programmable read-only memory
  • EPROM erasable and Programmable read only memory
  • PROM programmable read only memory
  • ROM read only memory
  • magnetic memory flash memory
  • flash memory magnetic disk or optical disk.
  • the power supply component 806 provides power for various components of the electronic device 800.
  • the power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.
  • the multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user.
  • the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user.
  • the touch panel includes one or more touch sensors to sense touch, sliding, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure related to the touch or slide operation.
  • the multimedia component 808 includes a front camera and/or a rear camera. When the electronic device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capabilities.
  • the audio component 810 is configured to output and/or input audio signals.
  • the audio component 810 includes a microphone (MIC), and when the electronic device 800 is in an operation mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal.
  • the received audio signal may be further stored in the memory 804 or transmitted via the communication component 816.
  • the audio component 810 further includes a speaker for outputting audio signals.
  • the I/O interface 812 provides an interface between the processing component 802 and a peripheral interface module.
  • the peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include, but are not limited to: home button, volume button, start button, and lock button.
  • the sensor component 814 includes one or more sensors for providing the electronic device 800 with various aspects of state evaluation.
  • the sensor component 814 can detect the on/off status of the electronic device 800 and the relative positioning of the components.
  • the component is the display and the keypad of the electronic device 800.
  • the sensor component 814 can also detect the electronic device 800 or the electronic device 800.
  • the position of the component changes, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration/deceleration of the electronic device 800, and the temperature change of the electronic device 800.
  • the sensor component 814 may include a proximity sensor configured to detect the presence of nearby objects when there is no physical contact.
  • the sensor component 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications.
  • the sensor component 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
  • the communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices.
  • the electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof.
  • the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel.
  • the communication component 816 further includes a near field communication (NFC) module to facilitate short-range communication.
  • the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
  • RFID radio frequency identification
  • IrDA infrared data association
  • UWB ultra-wideband
  • Bluetooth Bluetooth
  • the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field-available A programmable gate array (FPGA), controller, microcontroller, microprocessor, or other electronic components are implemented to implement the above methods.
  • ASIC application-specific integrated circuits
  • DSP digital signal processors
  • DSPD digital signal processing devices
  • PLD programmable logic devices
  • FPGA field-available A programmable gate array
  • controller microcontroller, microprocessor, or other electronic components are implemented to implement the above methods.
  • a non-volatile computer-readable storage medium such as the memory 804 including computer program instructions, which can be executed by the processor 820 of the electronic device 800 to complete the foregoing method.
  • the embodiments of the present disclosure also provide a computer program product, which includes computer-readable code.
  • the processor in the device executes the neural network training method provided in any of the above embodiments. Instructions.
  • the embodiments of the present disclosure also provide another computer program product for storing computer-readable instructions, which when executed, cause the computer to perform the operations of the image generation method provided by any of the foregoing embodiments.
  • the above-mentioned computer program product can be specifically implemented by hardware, software, or a combination thereof.
  • the computer program product is specifically embodied as a computer storage medium.
  • the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc. Wait.
  • SDK software development kit
  • Fig. 5 is a block diagram showing an electronic device 1900 according to an exemplary embodiment.
  • the electronic device 1900 may be provided as a server. 5
  • the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932, for storing instructions executable by the processing component 1922, such as application programs.
  • the application program stored in the memory 1932 may include one or more modules each corresponding to a set of instructions.
  • the processing component 1922 is configured to execute instructions to perform the above-described methods.
  • the electronic device 1900 may also include a power supply component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to the network, and an input output (I/O) interface 1958 .
  • the electronic device 1900 can operate based on an operating system stored in the memory 1932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
  • a non-volatile computer-readable storage medium is also provided, such as the memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the electronic device 1900 to complete the foregoing method.
  • the present disclosure may be a system, method and/or computer program product.
  • the computer program product may include a computer-readable storage medium loaded with computer-readable program instructions for enabling a processor to implement various aspects of the present disclosure.
  • the computer-readable storage medium may be a tangible device that can hold and store instructions used by the instruction execution device.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
  • Non-exhaustive list of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) Or flash memory), static random access memory (SRAM), portable compact disk read-only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, mechanical encoding device, such as a printer with instructions stored thereon
  • RAM random access memory
  • ROM read-only memory
  • EPROM erasable programmable read-only memory
  • flash memory flash memory
  • SRAM static random access memory
  • CD-ROM compact disk read-only memory
  • DVD digital versatile disk
  • memory stick floppy disk
  • mechanical encoding device such as a printer with instructions stored thereon
  • the computer-readable storage medium used here is not interpreted as the instantaneous signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (for example, light pulses through fiber optic cables), or through wires Transmission of electrical signals.
  • the computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to various computing/processing devices, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and/or a wireless network.
  • the network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge servers.
  • the network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network, and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing/processing device .
  • the computer program instructions used to perform the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or in one or more programming languages.
  • Source code or object code written in any combination, the programming language includes object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as "C" language or similar programming languages.
  • Computer-readable program instructions can be executed entirely on the user's computer, partly on the user's computer, executed as a stand-alone software package, partly on the user's computer and partly executed on a remote computer, or entirely on the remote computer or server carried out.
  • the remote computer can be connected to the user's computer through any kind of network-including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect to the user's computer) connection).
  • LAN local area network
  • WAN wide area network
  • an electronic circuit such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), can be customized by using the status information of the computer-readable program instructions.
  • FPGA field programmable gate array
  • PDA programmable logic array
  • the computer-readable program instructions are executed to realize various aspects of the present disclosure.
  • These computer-readable program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine that makes these instructions when executed by the processor of the computer or other programmable data processing device , A device that implements the functions/actions specified in one or more blocks in the flowcharts and/or block diagrams is produced. It is also possible to store these computer-readable program instructions in a computer-readable storage medium. These instructions make computers, programmable data processing apparatuses, and/or other devices work in a specific manner. Thus, the computer-readable medium storing the instructions includes An article of manufacture, which includes instructions for implementing various aspects of the functions/actions specified in one or more blocks in the flowcharts and/or block diagrams.
  • each block in the flowchart or block diagram may represent a module, program segment, or part of an instruction, and the module, program segment, or part of an instruction contains one or more components for realizing the specified logical function.
  • Executable instructions may also occur in a different order than the order marked in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, or they can sometimes be executed in the reverse order, depending on the functions involved.
  • each block in the block diagram and/or flowchart, and the combination of the blocks in the block diagram and/or flowchart can be implemented by a dedicated hardware-based system that performs the specified functions or actions Or it can be realized by a combination of dedicated hardware and computer instructions.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Multimedia (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Quality & Reliability (AREA)
  • Algebra (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

一种神经网络训练方法及装置和图像生成方法及装置，所述方法包括：将第一随机向量输入生成网络，获得第一生成图像（S11）；将所述第一生成图像和第一真实图像分别输入判别网络，分别获得所述第一生成图像的第一判别分布与第一真实图像的第二判别分布（S12）；根据所述第一判别分布、所述第二判别分布、预设的第一目标分布以及预设的第二目标分布，确定所述判别网络的第一网络损失（S13）；根据所述第一判别分布和所述第二判别分布，确定所述生成网络的第二网络损失（S14）；根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络（S15）。

Description

神经网络训练方法及装置和图像生成方法及装置
本公开要求在2019年9月27日提交中国专利局、申请号为201910927729.6、申请名称为“神经网络训练方法及装置和图像生成方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
技术领域
本公开涉及计算机技术领域，尤其涉及一种神经网络训练方法及装置和图像生成方法及装置。
背景技术
在相关技术中，生成对抗网络(Generative Adversarial Networks，GAN)由两个模块组成，分别为判别网络(Discriminator)和生成网络(Generator)。受零和博弈(zero-sum game)的启发，两个网络通过互相对抗的方式达到最佳生成效果。在训练过程中，判别器通过奖励真目标和惩罚假目标来学习区分真实图像数据和生成网络生成的仿真图像，生成器则通过逐步缩小判别器对假目标的惩罚，使得判别器无法区分真实图像与生成图像，两者互相博弈、进化，最终达到以假乱真的效果。
在相关技术中，生成对抗网络由判别网络输出的一个单一标量来描述输入图片的真实性，再使用该标量计算网络的损失，进而训练生成对抗网络。
发明内容
本公开提出了一种神经网络训练方法及装置和图像生成方法及装置。
根据本公开的一方面，提供了一种神经网络训练方法，包括：
将第一随机向量输入生成网络，获得第一生成图像；
将所述第一生成图像和第一真实图像分别输入判别网络，分别获得所述第一生成图像的第一判别分布与第一真实图像的第二判别分布，其中，所述第一判别分布表示所述第一生成图像的真实程度的概率分布，所述第二判别分布表示所述第一真实图像的真实程度的概率分布；
根据所述第一判别分布、所述第二判别分布、预设的第一目标分布以及预设的第二目标分布，确定所述判别网络的第一网络损失，其中，所述第一目标分布为生成图像的目标概率分布，所述第二目标分布为真实图像的目标概率分布；
根据所述第一判别分布和所述第二判别分布，确定所述生成网络的第二网络损失；
根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络。
根据本公开的实施例的神经网络训练方法，判别网络可针对输入图像输出判别分布，以概率分布的形式描述输入图像的真实性，可从颜色、纹理、比例、背景等维度描述输入图像为真实图像的概率，可从多个方面考量输入图像的真实性，减少信息丢失，为神经网络训练提供更全面的监督信息以及更准确的训练方向，提高训练精度，最终提高生成图像的质量，使得生成网络可适用于生成高清图像。并且，预设了生成图像的目标概率分布以及真实图像的目标概率分布来指导训练过程，在训练过程中引导使真实图像和生成图像接近各自的目标概率分布，增大真实图像和生成图像的区分度，增强判别网络区分真实图像和生成图像的能力，进而提升生成网络生成的图像的质量。
在一种可能的实现方式中，根据所述第一判别分布、所述第二判别分布、预设的第一目标分布以及预设的第二目标分布，确定所述判别网络的第一网络损失，包括：
根据所述第一判别分布和所述第一目标分布，确定所述第一生成图像的第一分布损失；
根据所述第二判别分布和所述第二目标分布，确定所述第一真实图像的第二分布损失；
根据所述第一分布损失和所述第二分布损失，确定所述第一网络损失。
通过这种方式，预设了生成图像的目标概率分布以及真实图像的目标概率分布来指导训练过程，并分别确定各自的分布损失，在训练过程中引导使真实图像和生成图像接近各自的目标概率分布，增大真实图像和生成图像的区分度，为判别网络提供了更准确的角度信息，为判别网络提供更准确的训练方向，增强判别网络区分真实图像和生成图像的能力，进而提升生成网络生成的图像的质量。
在一种可能的实现方式中，根据所述第一判别分布和所述第一目标分布，确定所述第一生成图像的第一分布损失，包括：
将所述第一判别分布映射到所述第一目标分布的支撑集，获得第一映射分布；
确定所述第一映射分布与所述第一目标分布的第一相对熵；
根据所述第一相对熵，确定所述第一分布损失。
在一种可能的实现方式中，根据所述第二判别分布和所述第二目标分布，确定所述第一真实图像的第二分布损失，包括：
将所述第二判别分布映射到所述第二目标分布的支撑集，获得第二映射分布；
确定所述第二映射分布与所述第二目标分布的第二相对熵；
根据所述第二相对熵，确定所述第二分布损失。
在一种可能的实现方式中，根据所述第一分布损失和所述第二分布损失，确定所述第一网络损失，包括：
对所述第一分布损失和所述第二分布损失进行加权求和处理，获得所述第一网络损失。
在一种可能的实现方式中，根据所述第一判别分布和所述第二判别分布，确定所述生成网络的第二网络损失，包括：
确定所述第一判别分布与所述第二判别分布的第三相对熵；
根据所述第三相对熵，确定所述第二网络损失。
通过这种方式，可通过减小第一判别分布与第二判别分布的差异的方式训练生成网络，使得判别网络性能提高的同时，促进生成网络的性能提高，从而生成逼真程度较高的生成图像，使得生成网络可适用于生成高清图像。
在一种可能的实现方式中，根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络，包括：
根据所述第一网络损失，调整所述判别网络的网络参数；
根据所述第二网络损失，调整所述生成网络的网络参数；
在所述判别网络和所述生成网络满足训练条件的情况下，获得训练后的所述生成网络和所述判别网络。
在一种可能的实现方式中，根据所述第一网络损失，调整所述判别网络的网络参数，包括：
将第二随机向量输入生成网络，获得第二生成图像；
根据所述第二生成图像对第二真实图像进行插值处理，获得插值图像；
将所述插值图像输入所述判别网络，获得所述插值图像的第三判别分布；
根据所述第三判别分布，确定所述判别网络的网络参数的梯度；
在所述梯度大于或等于梯度阈值的情况下，根据所述第三判别分布确定梯度惩罚参数；
根据所述第一网络损失和所述梯度惩罚参数，调整所述判别网络的网络参数。
通过这种方式，可通过检测判别网络的网络参数的梯度是否大于或等于梯度阈值，来限制判别网络在训练中的梯度下降速度，从而限制判别网络的训练进度，减少判别网络出现梯度消失的概率，从而可持续优化生成网络，提高生成网络的性能，使生成网络生成图像的逼真程度较高，且适用于生成高清图像。
在一种可能的实现方式中，根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络，包括：
将至少一个历史训练周期中输入生成网络的第一随机向量输入当前训练周期的生成网络，获得至少一个第三生成图像；
将与所述至少一个历史训练周期中输入生成网络的第一随机向量对应的第一生成图像、至少一个所述第三生成图像以及至少一个真实图像分别输入当前训练周期的判别网络，分别获得至少一个第一生成图像的第四判别分布、至少一个第三生成图像的第五判别分布和至少一个真实图像的第六判别分布；
根据所述第四判别分布、所述第五判别分布和所述第六判别分布确定当前训练周期的生成网络的训练进度参数；
在所述训练进度参数小于或等于训练进度阈值的情况下，停止调整所述判别网络的网络参数，仅 调整所述生成网络的网络参数。
通过这种方式，可通过检查判别网络和生成网络的训练进度，来限制判别网络在训练中的梯度下降速度，从而限制判别网络的训练进度，减少判别网络出现梯度消失的概率，从而可持续优化生成网络，提高生成网络的性能，使生成网络生成图像的逼真程度较高，且适用于生成高清图像。
在一种可能的实现方式中，根据所述第四判别分布、所述第五判别分布和所述第六判别分布确定当前训练周期的生成网络的训练进度参数，包括：
分别获取至少一个所述第四判别分布的第一期望值、至少一个所述第五判别分布的第二期望值以及至少一个所述第六判别分布的第三期望值；
分别获取所述至少一个所述第一期望值的第一平均值、至少一个所述第二期望值的第二平均值以及至少一个所述第三期望值的第三平均值；
确定所述第三平均值与所述第二平均值的第一差值以及所述第二平均值与所述第一平均值的第二差值；
将所述第一差值与所述第二差值的比值确定为所述当前训练周期的生成网络的训练进度参数。
根据本公开的一方面，提供了一种图像生成方法，包括：
获取第三随机向量；
将所述第三随机向量输入训练后获得的生成网络进行处理，获得目标图像。
根据本公开的一方面，提供了一种神经网络训练装置，包括：
生成模块，用于将第一随机向量输入生成网络，获得第一生成图像；
判别模块，用于将所述第一生成图像和第一真实图像分别输入判别网络，分别获得所述第一生成图像的第一判别分布与第一真实图像的第二判别分布，其中，所述第一判别分布表示所述第一生成图像的真实程度的概率分布，所述第二判别分布表示所述第一真实图像的真实程度的概率分布；
第一确定模块，用于根据所述第一判别分布、所述第二判别分布、预设的第一目标分布以及预设的第二目标分布，确定所述判别网络的第一网络损失，其中，所述第一目标分布为生成图像的目标概率分布，所述第二目标分布为真实图像的目标概率分布；
第二确定模块，用于根据所述第一判别分布和所述第二判别分布，确定所述生成网络的第二网络损失；
训练模块，用于根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络。
在一种可能的实现方式中，所述第一确定模块被进一步配置为：
根据所述第一判别分布和所述第一目标分布，确定所述第一生成图像的第一分布损失；
根据所述第二判别分布和所述第二目标分布，确定所述第一真实图像的第二分布损失；
根据所述第一分布损失和所述第二分布损失，确定所述第一网络损失。
在一种可能的实现方式中，所述第一确定模块被进一步配置为：
将所述第一判别分布映射到所述第一目标分布的支撑集，获得第一映射分布；
确定所述第一映射分布与所述第一目标分布的第一相对熵；
根据所述第一相对熵，确定所述第一分布损失。
在一种可能的实现方式中，所述第一确定模块被进一步配置为：
将所述第二判别分布映射到所述第二目标分布的支撑集，获得第二映射分布；
确定所述第二映射分布与所述第二目标分布的第二相对熵；
根据所述第二相对熵，确定所述第二分布损失。
在一种可能的实现方式中，所述第一确定模块被进一步配置为：
对所述第一分布损失和所述第二分布损失进行加权求和处理，获得所述第一网络损失。
在一种可能的实现方式中，所述第二确定模块被进一步配置为：
确定所述第一判别分布与所述第二判别分布的第三相对熵；
根据所述第三相对熵，确定所述第二网络损失。
在一种可能的实现方式中，所述训练模块被进一步配置为：
根据所述第一网络损失，调整所述判别网络的网络参数；
根据所述第二网络损失，调整所述生成网络的网络参数；
在所述判别网络和所述生成网络满足训练条件的情况下，获得训练后的所述生成网络和所述判别网络。
在一种可能的实现方式中，所述训练模块被进一步配置为：
将第二随机向量输入生成网络，获得第二生成图像；
根据所述第二生成图像对第二真实图像进行插值处理，获得插值图像；
将所述插值图像输入所述判别网络，获得所述插值图像的第三判别分布；
根据所述第三判别分布，确定所述判别网络的网络参数的梯度；
在所述梯度大于或等于梯度阈值的情况下，根据所述第三判别分布确定梯度惩罚参数；
根据所述第一网络损失和所述梯度惩罚参数，调整所述判别网络的网络参数。
在一种可能的实现方式中，所述训练模块被进一步配置为：
将至少一个历史训练周期中输入生成网络的第一随机向量输入当前训练周期的生成网络，获得至少一个第三生成图像；
将与所述至少一个历史训练周期中输入生成网络的第一随机向量对应的第一生成图像、至少一个所述第三生成图像以及至少一个真实图像分别输入当前训练周期的判别网络，分别获得至少一个第一生成图像的第四判别分布、至少一个第三生成图像的第五判别分布和至少一个真实图像的第六判别分布；
根据所述第四判别分布、所述第五判别分布和所述第六判别分布确定当前训练周期的生成网络的训练进度参数；
在所述训练进度参数小于或等于训练进度阈值的情况下，停止调整所述判别网络的网络参数，仅调整所述生成网络的网络参数。
在一种可能的实现方式中，所述训练模块被进一步配置为：
分别获取至少一个所述第四判别分布的第一期望值、至少一个所述第五判别分布的第二期望值以及至少一个所述第六判别分布的第三期望值；
分别获取所述至少一个所述第一期望值的第一平均值、至少一个所述第二期望值的第二平均值以及至少一个所述第三期望值的第三平均值；
确定所述第三平均值与所述第二平均值的第一差值以及所述第二平均值与所述第一平均值的第二差值；
将所述第一差值与所述第二差值的比值确定为所述当前训练周期的生成网络的训练进度参数。
根据本公开的一方面，提供了一种图像生成装置，其中，包括：
获取模块，用于获取第三随机向量；
获得模块，用于将所述第三随机向量输入训练后获得的生成网络进行处理，获得目标图像。
根据本公开的一方面，提供了一种电子设备，包括：
处理器；
用于存储处理器可执行指令的存储器；
其中，所述处理器被配置为：执行上述方法。
根据本公开的一方面，提供了一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现上述方法。
根据本公开的一方面，提供了一种计算机程序，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行用于执行上述方法。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，而非限制本公开。
根据下面参考附图对示例性实施例的详细说明，本公开的其它特征及方面将变得清楚。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，这些附图示出了符合本公开的实施例，并与说明书一起用于说明本公开的技术方案。
图1示出根据本公开实施例的神经网络训练方法的流程图；
图2示出根据本公开实施例的神经网络训练方法的应用示意图；
图3示出根据本公开实施例的神经网络训练装置的框图；
图4示出根据本公开实施例的电子装置的框图；
图5示出根据本公开实施例的电子装置的框图。
具体实施方式
以下将参考附图详细说明本公开的各种示例性实施例、特征和方面。附图中相同的附图标记表示功能相同或相似的元件。尽管在附图中示出了实施例的各种方面，但是除非特别指出，不必按比例绘制附图。
在这里专用的词“示例性”意为“用作例子、实施例或说明性”。这里作为“示例性”所说明的任何实施例不必解释为优于或好于其它实施例。
本文中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本文中术语“至少一种”表示多种中的任意一种或多种中的至少两种的任意组合，例如，包括A、B、C中的至少一种，可以表示包括从A、B和C构成的集合中选择的任意一个或多个元素。
另外，为了更好的说明本公开，在下文的具体实施方式中给出了众多的具体细节。本领域技术人员应当理解，没有某些具体细节，本公开同样可以实施。在一些实例中，对于本领域技术人员熟知的方法、手段、元件和电路未作详细描述，以便于凸显本公开的主旨。
图1示出根据本公开实施例的神经网络训练方法的流程图，如图1所示，所述方法包括：
在步骤S11中，将第一随机向量输入生成网络，获得第一生成图像；
在步骤S12中，将所述第一生成图像和第一真实图像分别输入判别网络，分别获得所述第一生成图像的第一判别分布与第一真实图像的第二判别分布，其中，所述第一判别分布表示所述第一生成图像的真实程度的概率分布，所述第二判别分布表示所述第一真实图像的真实程度的概率分布；
在步骤S13中，根据所述第一判别分布、所述第二判别分布、预设的第一目标分布以及预设的第二目标分布，确定所述判别网络的第一网络损失，其中，所述第一目标分布为生成图像的目标概率分布，所述第二目标分布为真实图像的目标概率分布；
在步骤S14中，根据所述第一判别分布和所述第二判别分布，确定所述生成网络的第二网络损失；
在步骤S15中，根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络。
根据本公开的实施例的神经网络训练方法，判别网络可针对输入图像输出判别分布，以概率分布的形式描述输入图像的真实性，可从颜色、纹理、比例、背景等维度描述输入图像为真实图像的概率，可从多个方面考量输入图像的真实性，减少信息丢失，为神经网络训练提供更全面的监督信息以及更准确的训练方向，提高训练精度，最终提高生成图像的质量，使得生成网络可适用于生成高清图像。并且，预设了生成图像的目标概率分布以及真实图像的目标概率分布来指导训练过程，在训练过程中引导使真实图像和生成图像接近各自的目标概率分布，增大真实图像和生成图像的区分度，增强判别网络区分真实图像和生成图像的能力，进而提升生成网络生成的图像的质量。
在一种可能的实现方式中，所述神经网络训练方法可以由终端设备或其它处理设备执行，其中，终端设备可以为用户设备(User Equipment，UE)、移动设备、用户终端、终端、蜂窝电话、无绳电话、个人数字处理(Personal Digital Assistant，PDA)、手持设备、计算设备、车载设备、可穿戴设备等。其它处理设备可为服务器或云端服务器等。在一些可能的实现方式中，该神经网络训练方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。
在一种可能的实现方式中，所述神经网络可以是由生成网络和判别网络组成的生成对抗网络。生成网络可以是卷积神经网络等深度学习神经网络，本公开对生成网络的类型和结构不做限制。判别网 络可以是卷积神经网络等深度学习神经网络，本公开对判别网络的类型和结构不做限制。生成网络可对随机向量进行处理，获得生成图像，随机向量可以是各元素为随机数的向量，可通过随机采样等方式获得。在步骤S11中，可通过随机采样等方式获得第一随机向量，生成网络可对第一随机向量进行卷积等处理，获得与第一随机向量对应的第一生成图像。第一随机向量是随机生成的向量，因此，第一生成图像为随机图像。
在一种可能的实现方式中，第一真实图像可以是任意的真实图像，例如，可以是图像获取装置(例如，相机、摄像头等)拍摄到的真实图像。在步骤S12中，可将第一真实图像和第一生成图像分别输入判别网络，分别获得第一生成图像的第一判别分布和第一真实图像的第二判别分布，第一判别分布和第二判别分布可以是向量形式的参数，例如，可以用向量的形式表示概率分布。第一判别分布可表示第一生成图像的真实程度，即，可通过第一判别分布来描述第一生成图像是真实图像的概率。第二判别分布可表示第一真实图像的真实程度，即，可通过第二判别分布来描述第一真实图像是真实图像的概率。以分布(如多维向量)的形式描述图像的真实性，可从颜色、纹理、比例、背景等多个方面考量图像的真实性，减少信息丢失，为训练提供准确的训练方向。
在一种可能的实现方式中，在步骤S13中，可预设真实图像的目标概率分布(即，第二目标分布)，以及生成图像的目标概率分布(即，第一目标分布)，在训练过程中，可根据真实图像的目标概率分布以及生成图像的目标概率分布分别确定生成图像对应的网络损失和真实图像对应的网络损失，并分别利用生成图像对应的网络损失和真实图像对应的网络损失调整判别网络的参数，使真实图像的第二判别分布接近第二目标分布，且与第一目标分布有显著差别，并使生成图像的第一判别分布接近第一目标分布，且与第二目标分布有显著差别，可增大真实图像和生成图像的区分度，增强判别网络区分真实图像和生成图像的能力，进而提升生成网络生成的图像的质量。
在示例中，可预设生成图像的锚(anchor)分布(即，第一目标分布)和真实图像的锚分布(即，第二目标分布)，表示生成图像的锚分布的向量与表示真实图像的锚分布的向量具有显著差异。例如，预设一个固定分布U，设定第一目标分布A1＝U，第二目标分布A2＝U+1。在训练过程中，可通过调整判别网络的网络参数，使得第一判别分布与生成图像的锚分布的差异缩小，在此过程中，第一判别分布与真实图像的锚分布的差异将增大。训练过程中，通过调整判别网络的网络参数，还使得第二判别分布与真实图像的锚分布的差异缩小，在此过程中，第二判别分布与生成图像的锚分布的差异将增大。即，对真实图像和生成图像分别预设锚分布，使真实图像和生成图像的分布差异增大，从而提升判别网络对真实图像和生成图像的区分能力。
在一种可能的实现方式中，步骤S13可包括：根据所述第一判别分布和所述第一目标分布，确定所述第一生成图像的第一分布损失；根据所述第二判别分布和所述第二目标分布，确定所述第一真实图像的第二分布损失；根据所述第一分布损失和所述第二分布损失，确定所述第一网络损失。
在示例中，第一目标分布为准确的概率分布，可确定第一目标分布和第一判别分布之间的差异，从而确定第一分布损失。
在一种可能的实现方式中，可根据第一判别分布和第一目标分布，确定第一生成图像对应的网络损失(即，第一分布损失)。其中，根据所述第一判别分布和所述第一目标分布，确定所述第一生成图像的第一分布损失，包括：将所述第一判别分布映射到所述第一目标分布的支撑集，获得第一映射分布；确定所述第一映射分布与所述第一目标分布的第一相对熵；根据所述第一相对熵，确定所述第一分布损失。
在一种可能的实现方式中，第一判别分布和第一目标分布的支撑集(所述支撑集为表示概率分布的分布范围的拓扑空间)可能不同，即，第一判别分布的分布范围与第一目标分布的分布范围不同。在分布范围不同时，比较两种概率分布的差异没有意义，因此，可将第一判别分布映射到第一目标分布的支撑集，或将第一目标分布映射到第一判别分布的支撑集，又或者将第一判别分布和第一目标分布映射到同一个支撑集，即，使得第一判别分布的分布范围与第一目标分布的分布范围相同，可在相同的分布范围中比较两种概率分布的差异。在示例中，可通过线性变换等方式，例如利用投影矩阵对第一判别分布进行投影处理，将第一判别分布映射到第一目标分布的支撑集，即，可对第一判别分布 的向量进行线性变换，变换后获得的向量即为映射到第一目标分布的支撑集后的第一映射分布。
在一种可能的实现方式中，可确定第一映射分布与第一目标分布的第一相对熵，即，KL(Kullback-Leibler)距离，所述第一相对熵可表示相同支撑集中的两个概率分布的差异(即，第一映射分布与第一目标分布的差异)。当然，在其他实施方式中，也可以通过JS散度(Jensen-Shannon divergence)或Wasserstein距离等其他方式确定第一映射分布与第一目标分布的差异。
在一种可能的实现方式中，可根据第一相对熵，确定第一分布损失(即，生成图像对应的网络损失)。在示例中，可将第一相对熵确定为所述第一分布损失，或对第一相对熵进行运算处理，例如，对第一相对熵进行加权、取对数、取指数等处理，获得所述第一分布损失。本公开对第一分布损失的确定方式不做限制。
在示例中，第二目标分布为准确的概率分布，可确定第二目标分布和第二判别分布之间的差异，从而确定第二分布损失。
在一种可能的实现方式中，可根据第二判别分布和第二目标分布，确定第一真实图像对应的网络损失(即，第二分布损失)。其中，根据所述第二判别分布和所述第二目标分布，确定所述第一真实图像的第二分布损失，包括：将所述第二判别分布映射到所述第二目标分布的支撑集，获得第二映射分布；确定所述第二映射分布与所述第二目标分布的第二相对熵；根据所述第二相对熵，确定所述第二分布损失。
在一种可能的实现方式中，第二判别分布和第二目标分布的支撑集(所述支撑集为表示概率分布的分布范围的拓扑空间)可能不同，即，第二判别分布的分布范围与第二目标分布的分布范围不同。可将第二判别分布映射到第二目标分布的支撑集，或将第二目标分布映射到第二判别分布的支撑集，又或者将第二判别分布和第二目标分布映射到同一个支撑集，使得第二判别分布的分布范围与第二目标分布的分布范围相同，可在相同的分布范围中比较两种概率分布的差异。在示例中，可通过线性变换等方式，例如利用投影矩阵对第二判别分布进行投影处理，将第二判别分布映射到第二目标分布的支撑集，即，可对第二判别分布的向量进行线性变换，变换后获得的向量即为映射到第二目标分布的支撑集后的第二映射分布。
在一种可能的实现方式中，可确定第二映射分布与第二目标分布的第二相对熵，所述第二相对熵可表示相同支撑集中的两个概率分布的差异(即，第二映射分布与第二目标分布的差异)。其中，第二相对熵的计算方法与第一相对熵类似，此处不再重复。当然，在其他实施方式中，也可以通过JS散度(Jensen-Shannon divergence)或Wasserstein距离等其他方式确定第二映射分布与第二目标分布的差异。
在一种可能的实现方式中，可根据第二相对熵，确定第二分布损失(即，生成图像对应的网络损失)。在示例中，可将第二相对熵确定为所述第二分布损失，或对第二相对熵进行运算处理，例如，对第二相对熵进行加权、取对数、取指数等处理，获得所述第二分布损失。本公开对第二分布损失的确定方式不做限制。
在一种可能的实现方式中，可根据第一生成图像的第一分布损失和第二生成图像的第二分布损失来确定第一网络损失。其中，根据所述第一分布损失和所述第二分布损失，确定所述第一网络损失，包括：对所述第一分布损失和所述第二分布损失进行加权求和处理，获得所述第一网络损失。在示例中，第一分布损失和第二分布损失的权重可相同，即，将第一分布损失和第二分布损失直接求和，可获得第一网络损失。或者，可第一分布损失和第二分布损失的权重可不同，即，将第一分布损失和第二分布损失分别乘以各自的权重后再进行求和，可获得第一网络损失。第一分布损失和第二分布损失的权重可以是预设的，本公开对第一分布损失和第二分布损失的权重不做限制。
通过这种方式，预设了生成图像的目标概率分布以及真实图像的目标概率分布来指导训练过程，并分别确定各自的分布损失，在训练过程中引导使真实图像和生成图像接近各自的目标概率分布，增大真实图像和生成图像的区分度，为判别网络提供了更准确的角度信息，为判别网络提供更准确的训练方向，增强判别网络区分真实图像和生成图像的能力，进而提升生成网络生成的图像的质量。
在一种可能的实现方式中，还可确定生成网络的第二网络损失。在示例中，判别网络需要判别输 入图像为真实图像还是生成图像，因此，判别网络在训练过程中可增强对真实图像和生成图像的区分能力，即，使真实图像和生成图像的判别分布接近各自的目标概率分布，从而增大真实图像和生成图像的区分度。然而，生成网络的目标为使生成图像接近真实图像，即，使生成图像足够逼真，使得判别网络难以辨别出生成网络输出的生成图像。在对抗训练达到平衡状态时，判别网络和生成网络的性能都较强，即，判别网络的判别能力很强，能够分辨出真实图像和逼真程度较低的生成图像，而生成网络生成的图像逼真程度很高，使判别网络难以分辨出高质量的生成图像。在对抗训练中，判别网络性能提升可促进生成网络的性能提升，即，判别网络分别真实图像和生成图像的能力越强，则会促使生成网络生成的图像逼真程度越高。
生成网络的训练目的为提高生成图像的逼真程度，即，使得生成图像接近真实图像。也就是说，生成网络的训练可以使第一生成图像的第一判别分布与第一真实图像的第二判别分布接近，从而使得判别网络难以辨别。在一种可能的实现方式中，步骤S14可包括：确定所述第一判别分布与所述第二判别分布的第三相对熵；根据所述第三相对熵，确定所述第二网络损失。
在一种可能的实现方式中，可确定第一判别分布与第二判别分布的第三相对熵，所述第三相对熵表示相同支撑集中的两个概率分布的差异(即，第三映射分布与第四映射分布的差异)。其中，第三相对熵的计算方法与第一相对熵类似，此处不再重复。当然，在其他实施方式中，也可以通过JS散度(Jensen-Shannon divergence)或Wasserstein距离等其他方式确定第一判别分布与第二判别分布的差异，以通过二者差异确定生成网络的网络损失。
在一种可能的实现方式中，可根据第三相对熵，确定第二网络损失。在示例中，可将第三相对熵确定为第二网络损失，或对第三相对熵进行运算处理，例如，对第三相对熵进行加权、取对数、取指数等处理，获得第二网络损失。本公开对第二网络损失的确定方式不做限制。
在一种可能的实现方式中，第一判别分布与第二判别分布的支撑集不同，即，第一判别分布与第二判别分布的分布范围可不同。可经过线性变换使第一判别分布与第二判别分布的支持集重合，例如，可将第一判别分布与第二判别分布映射到目标支撑集，使得第二判别分布的分布范围与第一判别分布的分布范围相同，可在相同的分布范围中比较两种概率分布的差异。
在示例中，所述目标支撑集是所述第一判别分布的支撑集或所述第二判别分布的支撑集。可通过线性变换等方式，将第二判别分布映射到第一判别分布的支撑集，即，可对第二判别分布的向量进行线性变换，变换后获得的向量即为映射到第一判别分布的支撑集后的第四映射分布，并将第一判别分布作为所述第三映射分布。或者，可通过线性变换等方式，将第一判别分布映射到第二判别分布的支撑集，即，可对第一判别分布的向量进行线性变换，变换后获得的向量即为映射到第二判别分布的支撑集后的第三映射分布，并将第二判别分布作为所述第四映射分布。
在示例中，所述目标支撑集也可以是其他支撑集，例如，可预设一支撑集，并将第一判别分布和第二判别分布均映射到该支撑集，分别获得第三映射分布和第四映射分布。进一步地，可计算第三映射分布和第四映射分布的第三相对熵。本公开对目标支撑集不做限制。
通过这种方式，可通过减小第一判别分布与第二判别分布的差异的方式训练生成网络，使得判别网络性能提高的同时，促进生成网络的性能提高，从而生成逼真程度较高的生成图像，使得生成网络可适用于生成高清图像。
在一种可能的实现方式中，可根据判别网络的第一网络损失和生成网络的第二网络损失，对抗训练生成网络和判别网络。即，通过训练，使生成网络和判别网络的性能同时提高，提高判别网络的分辨能力，且提高生成网络生成逼真度较高的生成图像的能力，且使生成网络和判别网络达到平衡状态。
可选地，步骤S15可包括：根据所述第一网络损失，调整所述判别网络的网络参数；根据所述第二网络损失，调整所述生成网络的网络参数；在所述判别网络和所述生成网络满足训练条件的情况下，获得训练后的所述生成网络和所述判别网络。
在训练过程中，由于网络参数的复杂程度不同等因素，判别网络的训练进度通常领先于生成网络，而如果判别网络进度较快，提前训练完成，则无法为生成网络提供反向传播中的梯度，进而无法更新生成网络的参数，即，无法提升生成网络的性能。因此，生成网络生成的图像的性能受到限制，不适 用于生成高清的图像，且逼真度较低。
在一种可能的实现方式中，可限制在判别网络的训练过程中，用于调整判别网络的网络参数的梯度。其中，根据所述第一网络损失，调整所述判别网络的网络参数，包括：将第二随机向量输入生成网络，获得第二生成图像；根据所述第二生成图像对第二真实图像进行插值处理，获得插值图像；将所述插值图像输入所述判别网络，获得所述插值图像的第三判别分布；根据所述第三判别分布，确定所述判别网络的网络参数的梯度；在所述梯度大于梯度阈值的情况下，根据所述第三判别分布确定梯度惩罚参数；根据所述第一网络损失和所述梯度惩罚参数，调整所述判别网络的网络参数。
在一种可能的实现方式中，可通过随机采样等方式获得第二随机向量，并输入生成网络，获得第二生成图像，即，获得一张非真实图像。也可通过其他方式获得第二生成图像，例如，可直接随机生成一张非真实图像。
在一种可能的实现方式中，可将第二生成图像和第二真实图像进行插值处理，获得插值图像，即，插值图像为真实图像与非真实图像的合成图像，在插值图像中，包括部分真实图像，也包括部分非真实图像。在示例中，可对第二真实图像和第二生成图像进行随机非线性插值，获得所述插值图像，本公开对插值图像的获得方式不做限制。
在一种可能的实现方式中，可将插值图像输入判别网络，获得插值图像的第三判别分布，即，判别网络可针对该真实图像与非真实图像的合成图像进行判别处理，获得第三判别分布。
在一种可能的实现方式中，可利用第三判别分布来确定判别网络的网络参数的梯度，例如，可预设插值图像的目标概率分布(例如，可表示插值图像为真实图像的概率为50％的目标概率分布)，并利用第三判别分布和目标概率分布的相对熵来确定判别网络的网络参数的梯度。例如，可将第三判别分布和目标概率分布的相对熵进行反向传播，计算该相对熵与判别网络的各网络参数的偏微分，从而获得网络参数的梯度。当然，在其他可能的实现方式中，也可以利用第三判别分布和目标概率分布的JS散度等其他类型的差异，确定判别网络的参数梯度。
在一种可能的实现方式中，如果判别网络的网络参数的梯度大于或等于预设的梯度阈值，则可根据第三判别分布确定梯度惩罚参数。梯度阈值可以是对梯度进行限制的阈值，如果梯度较大，则在训练过程中，梯度的下降速度可能较快(即，训练步长较大，网络损失趋于最小值的速度较快)，因此，可通过梯度阈值对梯度进行限制。在示例中，梯度阈值可设为10、20等，本公开对梯度阈值不做限制。
在示例中，通过梯度惩罚参数对超过梯度阈值的网络参数的梯度进行调整，或对梯度下降速度进行限制，使得该参数的梯度较平缓，梯度下降速度减慢。例如，可根据第三判别分布的期望值确定梯度惩罚参数。梯度惩罚参数可以是对梯度下降的补偿参数，例如，可通过梯度惩罚参数调整偏微分的乘数，或通过梯度惩罚参数改变梯度下降的方向，以对梯度进行限制，从而减小判别网络的网络参数的梯度下降速度，防止判别网络的梯度下降过快，造成判别网络过早收敛(即，过快训练完成)。在示例中，第三判别分布为概率分布，可计算该概率分布的期望值，并根据期望值确定所述梯度惩罚参数，例如，可将所述期望值确定为网络参数的偏微分的乘数，即，将期望值确定为梯度惩罚参数，并将梯度惩罚参数作为梯度的乘数，本公开对梯度惩罚参数的确定方式不做限制。
在一种可能的实现方式中，可根据第一网络损失和梯度惩罚参数，调整判别网络的网络参数。即，在对第一网络损失进行反向传播使得梯度下降的过程中，加入梯度惩罚参数，在调整判别网络的网络参数的同时，防止梯度下降过快，即，防止判别网络过早训练完成。例如，可将梯度惩罚参数作为偏微分的乘数，即梯度的乘数，以减缓梯度下降速度，防止判别网络过早训练完成。
在一种可能的实现方式中，如果判别网络的网络参数的梯度小于预设的梯度阈值，则可根据第一网络损失调整判别网络的网络参数，即，对第一网络损失进行反向传播使梯度下降，使得第一网络损失减小。
在一种可能的实现方式中，可在调整判别网络的网络参数时，对判别网络的梯度是否大于或等于梯度阈值进行检验，在判别网络的梯度大于或等于梯度阈值的情况下设置梯度惩罚参数。也可不检验判别网络的梯度，而通过其他方式控制判别网络的训练进度(例如，暂停判别网络的网络参数的调整，仅调整生成网络的网络参数等)。
通过这种方式，可通过检测判别网络的网络参数的梯度是否大于或等于梯度阈值，来限制判别网络在训练中的梯度下降速度，从而限制判别网络的训练进度，减少判别网络出现梯度消失的概率，从而可持续优化生成网络，提高生成网络的性能，使生成网络生成图像的逼真程度较高，且适用于生成高清图像。
在一种可能的实现方式中，可根据第二网络损失调整生成网络的网络参数，例如，对第二网络损失进行反向传播使梯度下降，使得第二网络损失减小，以提升生成网络的性能。
在一种可能的实现方式中，可对抗训练判别网络和生成网络，在通过第一网络损失调整判别网络的网络参数时，保持生成网络的网络参数保持不变，在通过第二网络损失调整生成网络的网络参数时，保持判别网络的网络参数保持不变。可迭代执行上述训练过程，直到判别网络和生成网络满足训练条件，在示例中，所述训练条件包括判别网络和生成网络达到平衡状态，例如，判别网络和生成网络的网络损失均小于或等于预设阈值，或收敛于预设区间。或者，所述训练条件包括以下两个条件达到平衡状态：第一，生成网络的网络损失小于或等于预设阈值或收敛于预设区间，第二，判别网络输出的判别分布表示的输入图像为真实图像的概率最大化。此时，判别网络分别真实图像和生成图像的能力较强，生成网络生成的图像质量较高，逼真度较高。
在一种可能的实现方式中，除检验判别网络的梯度是否大于或等于梯度阈值之外，还可通过控制判别网络的训练进度的方式，减小判别网络出现梯度消失的概率。
在一种可能的实现方式中，可在任意训练周期结束后，检查判别网络和生成网络的训练进度。具体地，步骤S15可包括：将至少一个历史训练周期中输入生成网络的第一随机向量输入当前训练周期的生成网络，获得至少一个第三生成图像；将与所述至少一个历史训练周期中输入生成网络的第一随机向量对应的第一生成图像、至少一个所述第三生成图像以及至少一个真实图像分别输入当前训练周期的判别网络，分别获得至少一个第一生成图像的第四判别分布、至少一个第三生成图像的第五判别分布和至少一个真实图像的第六判别分布；根据所述第四判别分布、所述第五判别分布和所述第六判别分布确定当前训练周期的生成网络的训练进度参数；在所述训练进度参数小于或等于训练进度阈值的情况下，停止调整所述判别网络的网络参数，仅调整所述生成网络的网络参数。
在一种可能的实现方式中，可在训练过程中开辟一个缓存区，例如，经验缓存区(experience buffer)，在该缓存区中，可保存至少一个(例如，M个，M为正整数)历史训练周期的第一随机向量以及上述M个历史训练周期中生成网络根据第一随机向量生成的M个第一生成图像，即，每个历史训练周期均可通过一个第一随机向量生成一个第一生成图像，在缓存区中，可保存M个历史训练周期的第一随机向量，以及生成的M个第一生成图像。随着训练的进行，在训练周期数超过M时，可使用最新的训练周期的第一随机向量和第一生成图像代替最早存入缓存区的第一随机向量和第一生成图像。
在一种可能的实现方式中，可将至少一个历史训练周期中输入生成网络的第一随机向量输入当前训练周期的生成网络，获得至少一个第三生成图像，例如，可将缓存区中的m(m小于或等于M，且m为正整数)个第一随机向量输入当前训练周期的生成网络，获得m个第三生成图像。
在一种可能的实现方式中，可通过当前训练周期的判别网络分别对m个第三生成图像进行判别处理，获得m个第五判别分布。可通过当前训练周期的判别网络分别对m个历史训练周期的第一生成图像进行判别处理，获得m个第四判别分布。并可从数据库中随机采样得到m个真实图像，并通过当前训练周期的判别网络分别对m个真实图像进行判别处理，获得m个第六判别分布。
在一种可能的实现方式中，可根据m个第四判别分布、m个第五判别分布和m个第六判别分布来确定当前训练周期的生成网络的训练进度参数，即，确定判别网络的训练进度是否显著领先于生成网络，并在确定显著领先的情况下，调整生成网络的训练进度参数，以提高生成网络的训练进度，降低判别网络和生成网络的训练进度差异，即，暂停判别网络的训练，单独训练生成网络，使生成网络的进度参数提高，进度加快。
在一种可能的实现方式中，根据所述第四判别分布、所述第五判别分布和所述第六判别分布确定当前训练周期的生成网络的训练进度参数，包括：分别获取至少一个所述第四判别分布的第一期望值、至少一个所述第五判别分布的第二期望值以及至少一个所述第六判别分布的第三期望值；分别获取所 述至少一个所述第一期望值的第一平均值、至少一个所述第二期望值的第二平均值以及至少一个所述第三期望值的第三平均值；确定所述第三平均值与所述第二平均值的第一差值以及所述第二平均值与所述第一平均值的第二差值；将所述第一差值与所述第二差值的比值确定为所述当前训练周期的生成网络的训练进度参数。
在一种可能的实现方式中，可分别计算m个第四判别分布的期望值，获得m个的第一期望值，可分别计算m个第五判别分布的期望值，获得m个的第二期望值，并分别计算m个第六判别分布的期望值，获得m个的第三期望值。进一步地，可对m个的第一期望值进行平均处理，获得第一平均值S B，可对m个的第二期望值进行平均处理，获得第二平均值S G，并可对m个的第三期望值进行平均处理，获得第三平均值S R。
在一种可能的实现方式中，可确定第三平均值与第二平均值的第一差值(S R-S G)，并确定第二平均值与第一平均值的第二差值(S G-S B)。进一步地，可将第一差值与第二差值的比值(S R-S G)/(S G-S B)确定为所述当前训练周期的生成网络的训练进度参数。在另一示例中，还可将预设训练次数作为生成网络的训练进度参数，例如，可使生成网络和判别网络每共同训练100次，暂停判别网络训练，并单独训练生成网络50次，之后再使生成网络和判别网络每共同训练100次……直到生成网络和判别网络满足训练条件。
在一种可能的实现方式中，可设定训练进度阈值，所述训练进度阈值为确定生成网络训练进度的阈值，如果训练进度参数小于或等于训练进度阈值，则表明判别网络的训练进度显著领先于生成网络，即，生成网络的训练进度较慢，可暂停调整判别网络的网络参数，仅调整生成网络的网络参数。在示例中，可在接下来的训练周期中，重复执行以上检查判别网络和生成网络的训练进度，直到训练进度参数大于训练进度阈值，则可同时调整判别网络和生成网络的网络参数，即，使判别网络的训练暂停至少一个训练周期，仅训练生成网络(即，仅根据第三网络损失调整生成网络的网络参数，保持判别网络的网络参数不变)，直到生成网络的训练进度接近判别网络的训练进度，再对抗训练生成网络和判别网络。
在其他实现方式中，也可以在训练进度参数小于或等于训练进度阈值的情况下，降低判别网络的训练速度，例如延长判别网络的训练周期或降低判别网络的梯度下降速度等，直到训练进度参数大于训练进度阈值，则可恢复判别网络的训练速度。
通过这种方式，可通过检查判别网络和生成网络的训练进度，来限制判别网络在训练中的梯度下降速度，从而限制判别网络的训练进度，减少判别网络出现梯度消失的概率，从而可持续优化生成网络，提高生成网络的性能，使生成网络生成图像的逼真程度较高，且适用于生成高清图像。
在一种可能的实现方式中，在生成网络和判别网络的对抗训练完成后，即，生成网络和判别网络的性能较好时，可使用生成网络生成图像，生成的图像逼真度较高。
本公开还提供一种图像生成方法，使用上述训练完成的生成对抗网络生成图像。
在本公开的一些实施例中，一种图像生成方法包括：获取第三随机向量；将第三随机向量输入上述神经网络训练方法训练后获得的生成网络进行处理，获得目标图像。
在示例中，可通过随机采样等方式获得第三随机向量，并将第三随机向量输入训练后的生成网络。生成网络可输出逼真度较高的目标图像。在示例中，所述目标图像可以是高清图像，即，训练后的生成网络可适用于生成逼真度较高的高清图像。
根据本公开的实施例的神经网络训练方法，判别网络可针对输入图像输出判别分布，以分布的形式描述输入图像的真实性，从多个方面考量输入图像的真实性，减少信息丢失，为神经网络训练提供更全面的监督信息以及更准确的训练方向，提高训练精度，提高生成图像的质量，使得生成网络可适用于生成高清图像。并且预设了生成图像的目标概率分布以及真实图像的目标概率分布来指导训练过程，并分别确定各自的分布损失，在训练过程中引导使真实图像和生成图像接近各自的目标概率分布，增大真实图像和生成图像的区分度，增强判别网络区分真实图像和生成图像的能力，并通过减小第一判别分布与第二判别分布的差异的方式训练生成网络，使得判别网络性能提高的同时，促进生成网络的性能提高，从而生成逼真程度较高的生成图像，使得生成网络可适用于生成高清图像。进一步地， 还可通过检测判别网络的网络参数的梯度是否大于或等于梯度阈值，或检查判别网络和生成网络的训练进度，来限制判别网络在训练中的梯度下降速度，从而限制判别网络的训练进度，减少判别网络出现梯度消失的概率，从而可持续优化生成网络，提高生成网络的性能，使生成网络生成图像的逼真程度较高，且适用于生成高清图像。
图2示出根据本公开实施例的神经网络训练方法的应用示意图，如图2所示，可将第一随机向量输入生成网络，生成网络可输出第一生成图像。判别网络可将第一生成图像和第一真实图像分别进行判别处理，分别获得第一生成图像的第一判别分布和第一真实图像的第二判别分布。
在一种可能的实现方式中，可预设生成图像的锚分布(即，第一目标分布)和真实图像的锚分布(即，第二目标分布)。可根据第一判别分布和第一目标分布，确定第一生成图像对应的第一分布损失。并可根据第二判别分布和第二目标分布，确定第一真实图像对应的第二分布损失。进一步地，可通过第一分布损失和第二分布损失确定判别网络的第一网络损失。
在一种可能的实现方式中，可通过第一判别分布和第二判别分布确定生成网络的第二网络损失。进一步地，可通过第一网络损失和第二网络损失对抗训练生成网络和判别网络。即，通过第一网络损失调整判别网络的网络参数，以及通过第二网络损失调整生成网络的网络参数。
在一种可能的实现方式中，判别网络的训练进度通常比生成网络更快，为降低判别网络提前训练完成导致梯度消失的概率，从而造成生成网络无法继续优化。可通过检测判别网络的梯度，来控制判别网络的训练进度，在示例中，可对一张真实图像和生成图像进行插值，并通过判别网络来确定该插值图像的第三判别分布，进而根据第三判别分布的期望值确定梯度惩罚参数，如果判别网络的梯度大于或等于预设的梯度阈值，为防止判别网络的梯度下降过快，造成判别网络过快训练完成，可在对第一网络损失进行反向传播使得梯度下降的过程中，加入梯度惩罚参数，以限制判别网络的梯度下降速度。
在一种可能的实现方式中，还可检查判别网络和生成网络的训练进度，例如，可将M个历史训练周期中输入生成网络的M个第一随机向量输入当前训练周期的生成网络，获得M个第三生成图像。并根据M个历史训练周期中生成的第一生成图像、M个第三生成图像和M个真实图像来确定当前训练周期的生成网络的训练进度参数。如果训练进度参数小于或等于训练进度阈值，则表明判别网络的训练进度显著领先于生成网络，可暂停调整判别网络的网络参数，仅调整生成网络的网络参数。并在接下来的训练周期中，重复执行以上检查判别网络和生成网络的训练进度，直到训练进度参数大于训练进度阈值，方可同时调整判别网络和生成网络的网络参数，即，使判别网络的训练暂停至少一个训练周期，仅训练生成网络。
在一种可能的实现方式中，在生成网络和判别网络的对抗训练完成后，可使用生成网络生成目标图像，目标图像可以是逼真度较的高清图像。
在一种可能的实现方式中，所述神经网络训练方法可增强生成对抗的稳定性和生成图像的质量和逼真度。可适用于游戏中场景的生成或合成、图像风格的迁移或转换，以及图像聚类等场景，本公开对所述神经网络训练方法的使用场景不做限制。
图3示出根据本公开实施例的神经网络训练装置的框图，如图3所示，所述装置包括：
生成模块11，用于将第一随机向量输入生成网络，获得第一生成图像；
判别模块12，用于将所述第一生成图像和第一真实图像分别输入判别网络，分别获得所述第一生成图像的第一判别分布与第一真实图像的第二判别分布，其中，所述第一判别分布表示所述第一生成图像的真实程度的概率分布，所述第二判别分布表示所述第一真实图像的真实程度的概率分布；
第一确定模块13，用于根据所述第一判别分布、所述第二判别分布、预设的第一目标分布以及预设的第二目标分布，确定所述判别网络的第一网络损失，其中，所述第一目标分布为生成图像的目标概率分布，所述第二目标分布为真实图像的目标概率分布；
第二确定模块14，用于根据所述第一判别分布和所述第二判别分布，确定所述生成网络的第二网络损失；
训练模块15，用于根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判 别网络。
在一种可能的实现方式中，所述第一确定模块被进一步配置为：
根据所述第一判别分布和所述第一目标分布，确定所述第一生成图像的第一分布损失；
根据所述第二判别分布和所述第二目标分布，确定所述第一真实图像的第二分布损失；
根据所述第一分布损失和所述第二分布损失，确定所述第一网络损失。
在一种可能的实现方式中，所述第一确定模块被进一步配置为：
将所述第一判别分布映射到所述第一目标分布的支撑集，获得第一映射分布；
确定所述第一映射分布与所述第一目标分布的第一相对熵；
根据所述第一相对熵，确定所述第一分布损失。
在一种可能的实现方式中，所述第一确定模块被进一步配置为：
将所述第二判别分布映射到所述第二目标分布的支撑集，获得第二映射分布；
确定所述第二映射分布与所述第二目标分布的第二相对熵；
根据所述第二相对熵，确定所述第二分布损失。
在一种可能的实现方式中，所述第一确定模块被进一步配置为：
对所述第一分布损失和所述第二分布损失进行加权求和处理，获得所述第一网络损失。
在一种可能的实现方式中，所述第二确定模块被进一步配置为：
确定所述第一判别分布与所述第二判别分布的第三相对熵；
根据所述第三相对熵，确定所述第二网络损失。
在一种可能的实现方式中，所述训练模块被进一步配置为：
根据所述第一网络损失，调整所述判别网络的网络参数；
根据所述第二网络损失，调整所述生成网络的网络参数；
在所述判别网络和所述生成网络满足训练条件的情况下，获得训练后的所述生成网络和所述判别网络。
在一种可能的实现方式中，所述训练模块被进一步配置为：
将第二随机向量输入生成网络，获得第二生成图像；
根据所述第二生成图像对第二真实图像进行插值处理，获得插值图像；
将所述插值图像输入所述判别网络，获得所述插值图像的第三判别分布；
根据所述第三判别分布，确定所述判别网络的网络参数的梯度；
在所述梯度大于或等于梯度阈值的情况下，根据所述第三判别分布确定梯度惩罚参数；
根据所述第一网络损失和所述梯度惩罚参数，调整所述判别网络的网络参数。
在一种可能的实现方式中，所述训练模块被进一步配置为：
将至少一个历史训练周期中输入生成网络的第一随机向量输入当前训练周期的生成网络，获得至少一个第三生成图像；
将与所述至少一个历史训练周期中输入生成网络的第一随机向量对应的第一生成图像、至少一个所述第三生成图像以及至少一个真实图像分别输入当前训练周期的判别网络，分别获得至少一个第一生成图像的第四判别分布、至少一个第三生成图像的第五判别分布和至少一个真实图像的第六判别分布；
根据所述第四判别分布、所述第五判别分布和所述第六判别分布确定当前训练周期的生成网络的训练进度参数；
在所述训练进度参数小于或等于训练进度阈值的情况下，停止调整所述判别网络的网络参数，仅调整所述生成网络的网络参数。
在一种可能的实现方式中，所述训练模块被进一步配置为：
分别获取至少一个所述第四判别分布的第一期望值、至少一个所述第五判别分布的第二期望值以及至少一个所述第六判别分布的第三期望值；
分别获取所述至少一个所述第一期望值的第一平均值、至少一个所述第二期望值的第二平均值以 及至少一个所述第三期望值的第三平均值；
确定所述第三平均值与所述第二平均值的第一差值以及所述第二平均值与所述第一平均值的第二差值；
将所述第一差值与所述第二差值的比值确定为所述当前训练周期的生成网络的训练进度参数。
本公开还提供一种图像生成装置，使用上述训练完成的生成对抗网络生成图像。
在本公开的一些实施例中，一种图像生成装置包括：
获取模块，用于获取第三随机向量；
获得模块，用于将所述第三随机向量输入训练后获得的生成网络进行处理，获得目标图像。
可以理解，本公开提及的上述各个方法实施例，在不违背原理逻辑的情况下，均可以彼此相互结合形成结合后的实施例，限于篇幅，本公开不再赘述。
此外，本公开还提供了神经网络训练装置、电子设备、计算机可读存储介质、程序，上述均可用来实现本公开提供的任一种神经网络训练方法，相应技术方案和描述和参见方法部分的相应记载，不再赘述。本领域技术人员可以理解，在具体实施方式的上述方法中，各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定，各步骤的具体执行顺序应当以其功能和可能的内在逻辑确定。在一些实施例中，本公开实施例提供的装置具有的功能或包含的模块可以用于执行上文方法实施例描述的方法，其具体实现可以参照上文方法实施例的描述，为了简洁，这里不再赘述。
本公开实施例还提出一种计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现上述方法。计算机可读存储介质可以是易失性计算机可读存储介质或非易失性计算机可读存储介质。
本公开实施例还提出一种电子设备，包括：处理器；用于存储处理器可执行指令的存储器；其中，所述处理器被配置为上述方法。电子设备可以被提供为终端、服务器或其它形态的设备。
图4是根据一示例性实施例示出的一种电子设备800的框图。例如，电子设备800可以是移动电话，计算机，数字广播终端，消息收发设备，游戏控制台，平板设备，医疗设备，健身设备，个人数字助理等终端。
参照图4，电子设备800可以包括以下一个或多个组件：处理组件802，存储器804，电源组件806，多媒体组件808，音频组件810，输入/输出(I/O)的接口812，传感器组件814，以及通信组件816。
处理组件802通常控制电子设备800的整体操作，诸如与显示，电话呼叫，数据通信，相机操作和记录操作相关联的操作。处理组件802可以包括一个或多个处理器820来执行指令，以完成上述的方法的全部或部分步骤。此外，处理组件802可以包括一个或多个模块，便于处理组件802和其他组件之间的交互。例如，处理组件802可以包括多媒体模块，以方便多媒体组件808和处理组件802之间的交互。
存储器804被配置为存储各种类型的数据以支持在电子设备800的操作。这些数据的示例包括用于在电子设备800上操作的任何应用程序或方法的指令，联系人数据，电话簿数据，消息，图片，视频等。存储器804可以由任何类型的易失性或非易失性存储设备或者它们的组合实现，如静态随机存取存储器(SRAM)，电可擦除可编程只读存储器(EEPROM)，可擦除可编程只读存储器(EPROM)，可编程只读存储器(PROM)，只读存储器(ROM)，磁存储器，快闪存储器，磁盘或光盘。
电源组件806为电子设备800的各种组件提供电力。电源组件806可以包括电源管理系统，一个或多个电源，及其他与为电子设备800生成、管理和分配电力相关联的组件。
多媒体组件808包括在所述电子设备800和用户之间的提供一个输出接口的屏幕。在一些实施例中，屏幕可以包括液晶显示器(LCD)和触摸面板(TP)。如果屏幕包括触摸面板，屏幕可以被实现为触摸屏，以接收来自用户的输入信号。触摸面板包括一个或多个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器可以不仅感测触摸或滑动动作的边界，而且还检测与所述触摸或滑动操作相关的持续时间和压力。在一些实施例中，多媒体组件808包括一个前置摄像头和/或后置摄像头。当电子设备800处于操作模式，如拍摄模式或视频模式时，前置摄像头和/或后置摄像头可以接收外部的多媒体数据。每个前置摄像头和后置摄像头可以是一个固定的光学透镜系统或具有焦距和光学变焦能力。
音频组件810被配置为输出和/或输入音频信号。例如，音频组件810包括一个麦克风(MIC)，当电子设备800处于操作模式，如呼叫模式、记录模式和语音识别模式时，麦克风被配置为接收外部音频信号。所接收的音频信号可以被进一步存储在存储器804或经由通信组件816发送。在一些实施例中，音频组件810还包括一个扬声器，用于输出音频信号。
I/O接口812为处理组件802和外围接口模块之间提供接口，上述外围接口模块可以是键盘，点击轮，按钮等。这些按钮可包括但不限于：主页按钮、音量按钮、启动按钮和锁定按钮。
传感器组件814包括一个或多个传感器，用于为电子设备800提供各个方面的状态评估。例如，传感器组件814可以检测到电子设备800的打开/关闭状态，组件的相对定位，例如所述组件为电子设备800的显示器和小键盘，传感器组件814还可以检测电子设备800或电子设备800一个组件的位置改变，用户与电子设备800接触的存在或不存在，电子设备800方位或加速/减速和电子设备800的温度变化。传感器组件814可以包括接近传感器，被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件814还可以包括光传感器，如CMOS或CCD图像传感器，用于在成像应用中使用。在一些实施例中，该传感器组件814还可以包括加速度传感器，陀螺仪传感器，磁传感器，压力传感器或温度传感器。
通信组件816被配置为便于电子设备800和其他设备之间有线或无线方式的通信。电子设备800可以接入基于通信标准的无线网络，如WiFi，2G或3G，或它们的组合。在一个示例性实施例中，通信组件816经由广播信道接收来自外部广播管理系统的广播信号或广播相关信息。在一个示例性实施例中，所述通信组件816还包括近场通信(NFC)模块，以促进短程通信。例如，在NFC模块可基于射频识别(RFID)技术，红外数据协会(IrDA)技术，超宽带(UWB)技术，蓝牙(BT)技术和其他技术来实现。
在示例性实施例中，电子设备800可以被一个或多个应用专用集成电路(ASIC)、数字信号处理器(DSP)、数字信号处理设备(DSPD)、可编程逻辑器件(PLD)、现场可编程门阵列(FPGA)、控制器、微控制器、微处理器或其他电子元件实现，用于执行上述方法。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器804，上述计算机程序指令可由电子设备800的处理器820执行以完成上述方法。
本公开实施例还提供了一种计算机程序产品，包括计算机可读代码，当计算机可读代码在设备上运行时，设备中的处理器执行用于实现如上任一实施例提供的神经网络训练方法的指令。
本公开实施例还提供了另一种计算机程序产品，用于存储计算机可读指令，指令被执行时使得计算机执行上述任一实施例提供的图像生成方法的操作。
上述计算机程序产品可以具体通过硬件、软件或其结合的方式实现。在一个可选实施例中，所述计算机程序产品具体体现为计算机存储介质，在另一个可选实施例中，计算机程序产品具体体现为软件产品，例如软件开发包(Software Development Kit，SDK)等等。
图5是根据一示例性实施例示出的一种电子设备1900的框图。例如，电子设备1900可以被提供为一服务器。参照图5，电子设备1900包括处理组件1922，其进一步包括一个或多个处理器，以及由存储器1932所代表的存储器资源，用于存储可由处理组件1922的执行的指令，例如应用程序。存储器1932中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理组件1922被配置为执行指令，以执行上述方法。
电子设备1900还可以包括一个电源组件1926被配置为执行电子设备1900的电源管理，一个有线或无线网络接口1950被配置为将电子设备1900连接到网络，和一个输入输出(I/O)接口1958。电子设备1900可以操作基于存储在存储器1932的操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM,LinuxTM，FreeBSDTM或类似。
在示例性实施例中，还提供了一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器1932，上述计算机程序指令可由电子设备1900的处理组件1922执行以完成上述方法。
本公开可以是系统、方法和/或计算机程序产品。计算机程序产品可以包括计算机可读存储介质，其上载有用于使处理器实现本公开的各个方面的计算机可读程序指令。
计算机可读存储介质可以是可以保持和存储由指令执行设备使用的指令的有形设备。计算机可读存储介质例如可以是――但不限于――电存储设备、磁存储设备、光存储设备、电磁存储设备、半导体存储设备或者上述的任意合适的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：便携式计算机盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、静态随机存取存储器(SRAM)、便携式压缩盘只读存储器(CD-ROM)、数字多功能盘(DVD)、记忆棒、软盘、机械编码设备、例如其上存储有指令的打孔卡或凹槽内凸起结构、以及上述的任意合适的组合。这里所使用的计算机可读存储介质不被解释为瞬时信号本身，诸如无线电波或者其他自由传播的电磁波、通过波导或其他传输媒介传播的电磁波(例如，通过光纤电缆的光脉冲)、或者通过电线传输的电信号。
这里所描述的计算机可读程序指令可以从计算机可读存储介质下载到各个计算/处理设备，或者通过网络、例如因特网、局域网、广域网和/或无线网下载到外部计算机或外部存储设备。网络可以包括铜传输电缆、光纤传输、无线传输、路由器、防火墙、交换机、网关计算机和/或边缘服务器。每个计算/处理设备中的网络适配卡或者网络接口从网络接收计算机可读程序指令，并转发该计算机可读程序指令，以供存储在各个计算/处理设备中的计算机可读存储介质中。
用于执行本公开操作的计算机程序指令可以是汇编指令、指令集架构(ISA)指令、机器指令、机器相关指令、微代码、固件指令、状态设置数据、或者以一种或多种编程语言的任意组合编写的源代码或目标代码，所述编程语言包括面向对象的编程语言—诸如Smalltalk、C++等，以及常规的过程式编程语言—诸如“C”语言或类似的编程语言。计算机可读程序指令可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络—包括局域网(LAN)或广域网(WAN)—连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。在一些实施例中，通过利用计算机可读程序指令的状态信息来个性化定制电子电路，例如可编程逻辑电路、现场可编程门阵列(FPGA)或可编程逻辑阵列(PLA)，该电子电路可以执行计算机可读程序指令，从而实现本公开的各个方面。
这里参照根据本公开实施例的方法、装置(系统)和计算机程序产品的流程图和/或框图描述了本公开的各个方面。应当理解，流程图和/或框图的每个方框以及流程图和/或框图中各方框的组合，都可以由计算机可读程序指令实现。
这些计算机可读程序指令可以提供给通用计算机、专用计算机或其它可编程数据处理装置的处理器，从而生产出一种机器，使得这些指令在通过计算机或其它可编程数据处理装置的处理器执行时，产生了实现流程图和/或框图中的一个或多个方框中规定的功能/动作的装置。也可以把这些计算机可读程序指令存储在计算机可读存储介质中，这些指令使得计算机、可编程数据处理装置和/或其他设备以特定方式工作，从而，存储有指令的计算机可读介质则包括一个制造品，其包括实现流程图和/或框图中的一个或多个方框中规定的功能/动作的各个方面的指令。
也可以把计算机可读程序指令加载到计算机、其它可编程数据处理装置、或其它设备上，使得在计算机、其它可编程数据处理装置或其它设备上执行一系列操作步骤，以产生计算机实现的过程，从而使得在计算机、其它可编程数据处理装置、或其它设备上执行的指令实现流程图和/或框图中的一个或多个方框中规定的功能/动作。
附图中的流程图和框图显示了根据本公开的多个实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或指令的一部分，所述模块、程序段或指令的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个连续的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或动作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
以上已经描述了本公开的各实施例，上述说明是示例性的，并非穷尽性的，并且也不限于所披露的各实施例。在不偏离所说明的各实施例的范围和精神的情况下，对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择，旨在最好地解释各实施例的原理、实际应用或对市场中的技术的技术改进，或者使本技术领域的其它普通技术人员能理解本文披露的各实施例。

Claims (25)

  1. 一种神经网络训练方法，其中，包括：
    将第一随机向量输入生成网络，获得第一生成图像；
    将所述第一生成图像和第一真实图像分别输入判别网络，分别获得所述第一生成图像的第一判别分布与第一真实图像的第二判别分布，其中，所述第一判别分布表示所述第一生成图像的真实程度的概率分布，所述第二判别分布表示所述第一真实图像的真实程度的概率分布；
    根据所述第一判别分布、所述第二判别分布、预设的第一目标分布以及预设的第二目标分布，确定所述判别网络的第一网络损失，其中，所述第一目标分布为生成图像的目标概率分布，所述第二目标分布为真实图像的目标概率分布；
    根据所述第一判别分布和所述第二判别分布，确定所述生成网络的第二网络损失；
    根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络。
  2. 根据权利要求1所述的方法，其中，根据所述第一判别分布、所述第二判别分布、预设的第一目标分布以及预设的第二目标分布，确定所述判别网络的第一网络损失，包括：
    根据所述第一判别分布和所述第一目标分布，确定所述第一生成图像的第一分布损失；
    根据所述第二判别分布和所述第二目标分布，确定所述第一真实图像的第二分布损失；
    根据所述第一分布损失和所述第二分布损失，确定所述第一网络损失。
  3. 根据权利要求2所述的方法，其中，根据所述第一判别分布和所述第一目标分布，确定所述第一生成图像的第一分布损失，包括：
    将所述第一判别分布映射到所述第一目标分布的支撑集，获得第一映射分布；
    确定所述第一映射分布与所述第一目标分布的第一相对熵；
    根据所述第一相对熵，确定所述第一分布损失。
  4. 根据权利要求2所述的方法，其中，根据所述第二判别分布和所述第二目标分布，确定所述第一真实图像的第二分布损失，包括：
    将所述第二判别分布映射到所述第二目标分布的支撑集，获得第二映射分布；
    确定所述第二映射分布与所述第二目标分布的第二相对熵；
    根据所述第二相对熵，确定所述第二分布损失。
  5. 根据权利要求2所述的方法，其中，根据所述第一分布损失和所述第二分布损失，确定所述第一网络损失，包括：
    对所述第一分布损失和所述第二分布损失进行加权求和处理，获得所述第一网络损失。
  6. 根据权利要求1-5中任一项所述的方法，其中，根据所述第一判别分布和所述第二判别分布，确定所述生成网络的第二网络损失，包括：
    确定所述第一判别分布与所述第二判别分布的第三相对熵；
    根据所述第三相对熵，确定所述第二网络损失。
  7. 根据权利要求1-6中任一项所述的方法，其中，根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络，包括：
    根据所述第一网络损失，调整所述判别网络的网络参数；
    根据所述第二网络损失，调整所述生成网络的网络参数；
    在所述判别网络和所述生成网络满足训练条件的情况下，获得训练后的所述生成网络和所述判别网络。
  8. 根据权利要求7所述的方法，其中，根据所述第一网络损失，调整所述判别网络的网络参数，包括：
    将第二随机向量输入生成网络，获得第二生成图像；
    根据所述第二生成图像对第二真实图像进行插值处理，获得插值图像；
    将所述插值图像输入所述判别网络，获得所述插值图像的第三判别分布；
    根据所述第三判别分布，确定所述判别网络的网络参数的梯度；
    在所述梯度大于或等于梯度阈值的情况下，根据所述第三判别分布确定梯度惩罚参数；
    根据所述第一网络损失和所述梯度惩罚参数，调整所述判别网络的网络参数。
  9. 根据权利要求1-8中任一项所述的方法，其中，根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络，包括：
    将至少一个历史训练周期中输入生成网络的第一随机向量输入当前训练周期的生成网络，获得至少一个第三生成图像；
    将与所述至少一个历史训练周期中输入生成网络的第一随机向量对应的第一生成图像、至少一个所述第三生成图像以及至少一个真实图像分别输入当前训练周期的判别网络，分别获得至少一个第一生成图像的第四判别分布、至少一个第三生成图像的第五判别分布和至少一个真实图像的第六判别分布；
    根据所述第四判别分布、所述第五判别分布和所述第六判别分布确定当前训练周期的生成网络的训练进度参数；
    在所述训练进度参数小于或等于训练进度阈值的情况下，停止调整所述判别网络的网络参数，仅调整所述生成网络的网络参数。
  10. 根据权利要求9所述的方法，其中，根据所述第四判别分布、所述第五判别分布和所述第六判别分布确定当前训练周期的生成网络的训练进度参数，包括：
    分别获取至少一个所述第四判别分布的第一期望值、至少一个所述第五判别分布的第二期望值以及至少一个所述第六判别分布的第三期望值；
    分别获取所述至少一个所述第一期望值的第一平均值、至少一个所述第二期望值的第二平均值以及至少一个所述第三期望值的第三平均值；
    确定所述第三平均值与所述第二平均值的第一差值以及所述第二平均值与所述第一平均值的第二差值；
    将所述第一差值与所述第二差值的比值确定为所述当前训练周期的生成网络的训练进度参数。
  11. 一种图像生成方法，其中，包括：
    获取第三随机向量；
    将所述第三随机向量输入根据权利要求1-10中任一项所述的方法训练后获得的生成网络进行处理，获得目标图像。
  12. 一种神经网络训练装置，其中，包括：
    生成模块，用于将第一随机向量输入生成网络，获得第一生成图像；
    判别模块，用于将所述第一生成图像和第一真实图像分别输入判别网络，分别获得所述第一生成图像的第一判别分布与第一真实图像的第二判别分布，其中，所述第一判别分布表示所述第一生成图像的真实程度的概率分布，所述第二判别分布表示所述第一真实图像的真实程度的概率分布；
    第一确定模块，用于根据所述第一判别分布、所述第二判别分布、预设的第一目标分布以及预设的第二目标分布，确定所述判别网络的第一网络损失，其中，所述第一目标分布为生成图像的目标概率分布，所述第二目标分布为真实图像的目标概率分布；
    第二确定模块，用于根据所述第一判别分布和所述第二判别分布，确定所述生成网络的第二网络损失；
    训练模块，用于根据所述第一网络损失和所述第二网络损失，对抗训练所述生成网络和所述判别网络。
  13. 根据权利要求12所述的装置，其中，所述第一确定模块被进一步配置为：
    根据所述第一判别分布和所述第一目标分布，确定所述第一生成图像的第一分布损失；
    根据所述第二判别分布和所述第二目标分布，确定所述第一真实图像的第二分布损失；
    根据所述第一分布损失和所述第二分布损失，确定所述第一网络损失。
  14. 根据权利要求13所述的装置，其中，所述第一确定模块被进一步配置为：
    将所述第一判别分布映射到所述第一目标分布的支撑集，获得第一映射分布；
    确定所述第一映射分布与所述第一目标分布的第一相对熵；
    根据所述第一相对熵，确定所述第一分布损失。
  15. 根据权利要求13所述的装置，其中，所述第一确定模块被进一步配置为：
    将所述第二判别分布映射到所述第二目标分布的支撑集，获得第二映射分布；
    确定所述第二映射分布与所述第二目标分布的第二相对熵；
    根据所述第二相对熵，确定所述第二分布损失。
  16. 根据权利要求13所述的装置，其中，所述第一确定模块被进一步配置为：
    对所述第一分布损失和所述第二分布损失进行加权求和处理，获得所述第一网络损失。
  17. 根据权利要求12-16中任一项所述的装置，其中，所述第二确定模块被进一步配置为：
    确定所述第一判别分布与所述第二判别分布的第三相对熵；
    根据所述第三相对熵，确定所述第二网络损失。
  18. 根据权利要求12-17中任一项所述的装置，其中，所述训练模块被进一步配置为：
    根据所述第一网络损失，调整所述判别网络的网络参数；
    根据所述第二网络损失，调整所述生成网络的网络参数；
    在所述判别网络和所述生成网络满足训练条件的情况下，获得训练后的所述生成网络和所述判别网络。
  19. 根据权利要求18所述的装置，其中，所述训练模块被进一步配置为：
    将第二随机向量输入生成网络，获得第二生成图像；
    根据所述第二生成图像对第二真实图像进行插值处理，获得插值图像；
    将所述插值图像输入所述判别网络，获得所述插值图像的第三判别分布；
    根据所述第三判别分布，确定所述判别网络的网络参数的梯度；
    在所述梯度大于或等于梯度阈值的情况下，根据所述第三判别分布确定梯度惩罚参数；
    根据所述第一网络损失和所述梯度惩罚参数，调整所述判别网络的网络参数。
  20. 根据权利要求11-19中任一项所述的装置，其中，所述训练模块被进一步配置为：
    将至少一个历史训练周期中输入生成网络的第一随机向量输入当前训练周期的生成网络，获得至少一个第三生成图像；
    将与所述至少一个历史训练周期中输入生成网络的第一随机向量对应的第一生成图像、至少一个所述第三生成图像以及至少一个真实图像分别输入当前训练周期的判别网络，分别获得至少一个第一生成图像的第四判别分布、至少一个第三生成图像的第五判别分布和至少一个真实图像的第六判别分布；
    根据所述第四判别分布、所述第五判别分布和所述第六判别分布确定当前训练周期的生成网络的训练进度参数；
    在所述训练进度参数小于或等于训练进度阈值的情况下，停止调整所述判别网络的网络参数，仅调整所述生成网络的网络参数。
  21. 根据权利要求20所述的装置，其中，所述训练模块被进一步配置为：
    分别获取至少一个所述第四判别分布的第一期望值、至少一个所述第五判别分布的第二期望值以及至少一个所述第六判别分布的第三期望值；
    分别获取所述至少一个所述第一期望值的第一平均值、至少一个所述第二期望值的第二平均值以及至少一个所述第三期望值的第三平均值；
    确定所述第三平均值与所述第二平均值的第一差值以及所述第二平均值与所述第一平均值的第二差值；
    将所述第一差值与所述第二差值的比值确定为所述当前训练周期的生成网络的训练进度参数。
  22. 一种图像生成装置，其中，包括：
    获取模块，用于获取第三随机向量；
    获得模块，用于将所述第三随机向量输入根据权利要求12-21中任一项所述的装置训练后获得的生成网络进行处理，获得目标图像。
  23. 一种电子设备，其中，包括：
    处理器；
    用于存储处理器可执行指令的存储器；
    其中，所述处理器被配置为：执行权利要求1至11中任意一项所述的方法。
  24. 一种计算机可读存储介质，其上存储有计算机程序指令，其中，所述计算机程序指令被处理器执行时实现权利要求1至11中任意一项所述的方法。
  25. 一种计算机程序，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行用于实现权利要求1-11中的任一权利要求所述的方法。
PCT/CN2019/124541 2019-09-27 2019-12-11 神经网络训练方法及装置和图像生成方法及装置 WO2021056843A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
SG11202103479VA SG11202103479VA (en) 2019-09-27 2019-12-11 Method and apparatus for neutral network training and method and apparatus for image generation
KR1020217010144A KR20210055747A (ko) 2019-09-27 2019-12-11 신경 네트워크 훈련 방법 및 장치, 이미지 생성 방법 및 장치
JP2021518079A JP7165818B2 (ja) 2019-09-27 2019-12-11 ニューラルネットワークのトレーニング方法及び装置並びに画像生成方法及び装置
US17/221,096 US20210224607A1 (en) 2019-09-27 2021-04-02 Method and apparatus for neutral network training, method and apparatus for image generation, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910927729.6 2019-09-27
CN201910927729.6A CN110634167B (zh) 2019-09-27 2019-09-27 神经网络训练方法及装置和图像生成方法及装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/221,096 Continuation US20210224607A1 (en) 2019-09-27 2021-04-02 Method and apparatus for neutral network training, method and apparatus for image generation, and storage medium

Publications (1)

Publication Number Publication Date
WO2021056843A1 true WO2021056843A1 (zh) 2021-04-01

Family

ID=68973281

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/124541 WO2021056843A1 (zh) 2019-09-27 2019-12-11 神经网络训练方法及装置和图像生成方法及装置

Country Status (7)

Country Link
US (1) US20210224607A1 (zh)
JP (1) JP7165818B2 (zh)
KR (1) KR20210055747A (zh)
CN (1) CN110634167B (zh)
SG (1) SG11202103479VA (zh)
TW (1) TWI752405B (zh)
WO (1) WO2021056843A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114881884A (zh) * 2022-05-24 2022-08-09 河南科技大学 一种基于生成对抗网络的红外目标样本增强方法

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2594070B (en) * 2020-04-15 2023-02-08 James Hoyle Benjamin Signal processing system and method
US11272097B2 (en) * 2020-07-30 2022-03-08 Steven Brian Demers Aesthetic learning methods and apparatus for automating image capture device controls
KR102354181B1 (ko) * 2020-12-31 2022-01-21 주식회사 나인티나인 비쥬얼라이징 구현 가능한 건설 사업 정보 관리 시스템 및 이의 제어 방법
CN112990211B (zh) * 2021-01-29 2023-07-11 华为技术有限公司 一种神经网络的训练方法、图像处理方法以及装置
TWI766690B (zh) * 2021-05-18 2022-06-01 詮隼科技股份有限公司 封包產生方法及封包產生系統之設定方法
KR102636866B1 (ko) * 2021-06-14 2024-02-14 아주대학교산학협력단 공간 분포를 이용한 휴먼 파싱 방법 및 장치
CN114501164A (zh) * 2021-12-28 2022-05-13 海信视像科技股份有限公司 音视频数据的标注方法、装置及电子设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107293289A (zh) * 2017-06-13 2017-10-24 南京医科大学 一种基于深度卷积生成对抗网络的语音生成方法
CN109377448A (zh) * 2018-05-20 2019-02-22 北京工业大学 一种基于生成对抗网络的人脸图像修复方法
CN109377452A (zh) * 2018-08-31 2019-02-22 西安电子科技大学 基于vae和生成式对抗网络的人脸图像修复方法
CN109919921A (zh) * 2019-02-25 2019-06-21 天津大学 基于生成对抗网络的环境影响程度建模方法
US20190228268A1 (en) * 2016-09-14 2019-07-25 Konica Minolta Laboratory U.S.A., Inc. Method and system for cell image segmentation using multi-stage convolutional neural networks

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100996209B1 (ko) * 2008-12-23 2010-11-24 중앙대학교 산학협력단 변화값 템플릿을 이용한 객체 모델링 방법 및 그 시스템
US8520958B2 (en) * 2009-12-21 2013-08-27 Stmicroelectronics International N.V. Parallelization of variable length decoding
JP6318211B2 (ja) * 2016-10-03 2018-04-25 株式会社Ｐｒｅｆｅｒｒｅｄ Ｎｅｔｗｏｒｋｓ データ圧縮装置、データ再現装置、データ圧縮方法、データ再現方法及びデータ転送方法
EP3336800B1 (de) * 2016-12-19 2019-08-28 Siemens Healthcare GmbH Bestimmen einer trainingsfunktion zum generieren von annotierten trainingsbildern
US10665326B2 (en) * 2017-07-25 2020-05-26 Insilico Medicine Ip Limited Deep proteome markers of human biological aging and methods of determining a biological aging clock
CN108495110B (zh) * 2018-01-19 2020-03-17 天津大学 一种基于生成式对抗网络的虚拟视点图像生成方法
CN108510435A (zh) * 2018-03-28 2018-09-07 北京市商汤科技开发有限公司 图像处理方法及装置、电子设备和存储介质
CN108615073B (zh) * 2018-04-28 2020-11-03 京东数字科技控股有限公司 图像处理方法及装置、计算机可读存储介质、电子设备
CN108805833B (zh) * 2018-05-29 2019-06-18 西安理工大学 基于条件对抗网络的字帖二值化背景噪声杂点去除方法
CN109933677A (zh) * 2019-02-14 2019-06-25 厦门一品威客网络科技股份有限公司 图像生成方法和图像生成系统
CN109920016B (zh) * 2019-03-18 2021-06-25 北京市商汤科技开发有限公司 图像生成方法及装置、电子设备和存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190228268A1 (en) * 2016-09-14 2019-07-25 Konica Minolta Laboratory U.S.A., Inc. Method and system for cell image segmentation using multi-stage convolutional neural networks
CN107293289A (zh) * 2017-06-13 2017-10-24 南京医科大学 一种基于深度卷积生成对抗网络的语音生成方法
CN109377448A (zh) * 2018-05-20 2019-02-22 北京工业大学 一种基于生成对抗网络的人脸图像修复方法
CN109377452A (zh) * 2018-08-31 2019-02-22 西安电子科技大学 基于vae和生成式对抗网络的人脸图像修复方法
CN109919921A (zh) * 2019-02-25 2019-06-21 天津大学 基于生成对抗网络的环境影响程度建模方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114881884A (zh) * 2022-05-24 2022-08-09 河南科技大学 一种基于生成对抗网络的红外目标样本增强方法
CN114881884B (zh) * 2022-05-24 2024-03-29 河南科技大学 一种基于生成对抗网络的红外目标样本增强方法

Also Published As

Publication number Publication date
JP2022504071A (ja) 2022-01-13
JP7165818B2 (ja) 2022-11-04
CN110634167B (zh) 2021-07-20
TW202113752A (zh) 2021-04-01
KR20210055747A (ko) 2021-05-17
CN110634167A (zh) 2019-12-31
SG11202103479VA (en) 2021-05-28
US20210224607A1 (en) 2021-07-22
TWI752405B (zh) 2022-01-11

Similar Documents

Publication Publication Date Title
WO2021056843A1 (zh) 神经网络训练方法及装置和图像生成方法及装置
TWI717923B (zh) 面部識別方法及裝置、電子設備和儲存介質
WO2020192252A1 (zh) 图像生成方法及装置、电子设备和存储介质
TWI747325B (zh) 目標對象匹配方法及目標對象匹配裝置、電子設備和電腦可讀儲存媒介
WO2021051650A1 (zh) 人脸和人手关联检测方法及装置、电子设备和存储介质
TWI736179B (zh) 圖像處理方法、電子設備和電腦可讀儲存介質
US20210012143A1 (en) Key Point Detection Method and Apparatus, and Storage Medium
WO2021051949A1 (zh) 一种图像处理方法及装置、电子设备和存储介质
CN105335684B (zh) 人脸检测方法及装置
WO2021139120A1 (zh) 网络训练方法及装置、图像生成方法及装置
CN109165738B (zh) 神经网络模型的优化方法及装置、电子设备和存储介质
US11734804B2 (en) Face image processing method and apparatus, electronic device, and storage medium
TW202105202A (zh) 影片處理方法及裝置、電子設備、儲存媒體和電腦程式
CN110909815A (zh) 神经网络训练、图像处理方法、装置及电子设备
TWI735112B (zh) 圖像生成方法、電子設備和儲存介質
TW202032425A (zh) 圖像處理方法及裝置、電子設備和儲存介質
EP3657497A1 (en) Method and device for selecting target beam data from a plurality of beams
WO2021036013A1 (zh) 检测器的配置方法及装置、电子设备和存储介质
CN112598063A (zh) 神经网络生成方法及装置、电子设备和存储介质
CN109698794A (zh) 一种拥塞控制方法、装置、电子设备及存储介质
CN110135349A (zh) 识别方法、装置、设备及存储介质
CN109447258B (zh) 神经网络模型的优化方法及装置、电子设备和存储介质
WO2021082381A1 (zh) 人脸识别方法及装置、电子设备和存储介质
WO2020224448A1 (zh) 交互方法及装置、音箱、电子设备和存储介质
WO2016041315A1 (zh) Pwm数据的处理方法及装置

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2021518079

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20217010144

Country of ref document: KR

Kind code of ref document: A

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19947310

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19947310

Country of ref document: EP

Kind code of ref document: A1