WO2020006963A1 - 生成图像检测模型的方法和装置 - Google Patents

生成图像检测模型的方法和装置 Download PDF

Info

Publication number
WO2020006963A1
WO2020006963A1 PCT/CN2018/116337 CN2018116337W WO2020006963A1 WO 2020006963 A1 WO2020006963 A1 WO 2020006963A1 CN 2018116337 W CN2018116337 W CN 2018116337W WO 2020006963 A1 WO2020006963 A1 WO 2020006963A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
sample
human body
body part
training
Prior art date
Application number
PCT/CN2018/116337
Other languages
English (en)
French (fr)
Inventor
徐珍琦
朱延东
王长虎
Original Assignee
北京字节跳动网络技术有限公司
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 北京字节跳动网络技术有限公司 filed Critical 北京字节跳动网络技术有限公司
Publication of WO2020006963A1 publication Critical patent/WO2020006963A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • 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

Definitions

  • Embodiments of the present application relate to the field of computer technology, and in particular, to a method and an apparatus for generating an image detection model.
  • the embodiments of the present application provide a method and a device for generating an image detection model.
  • an embodiment of the present application provides a method for generating an image detection model.
  • the method includes: obtaining a first training sample set, where the training sample includes a sample image and annotation information used to characterize whether the sample image is a first category image.
  • a first-class image detection model is trained by using a machine learning method.
  • intercepting an image of a preset human body part included in a sample image of the training sample to obtain a new sample image includes: inputting the sample image into a pre-trained human body part detection model, obtaining detection result information, detecting The result information includes position information of an image of a preset human body part included in the sample image, where the human body part detection model is used to characterize the correspondence between the image and the position information of the image of the preset human body part included in the image; based on the obtained The position information is used to intercept the sample image to obtain a new sample image.
  • the detection result information further includes: category information and confidence of the preset human body part displayed in the image of the preset human body part included in the sample image.
  • the sample image is intercepted based on the obtained position information to obtain a new sample image, including: in order of increasing confidence, based on the obtained position information, a preset is intercepted from the sample image. The number of images of the preset human body part, and the captured image of the preset human body part as a new sample image.
  • the human body part detection model is obtained by training in the following steps: obtaining a second training sample set, where the training sample includes sample images and label information of the sample images, wherein the label information includes preset human body parts included in the sample image; The position information of the image and the category information of the preset human body part displayed in the sample image; the sample image of the training sample in the second training sample set is used as input, and the label information corresponding to the input sample image is used as the desired output, and the training is Human body part detection model.
  • an embodiment of the present application provides an image detection method.
  • the method includes: acquiring an image to be detected; inputting the image to be detected into a first-type image detection model to obtain whether the image to be detected is a first-type image; Information of the detection result, wherein the image detection model of the first category is generated according to a method described in any implementation manner of the first aspect.
  • an embodiment of the present application provides an apparatus for generating an image detection model.
  • the apparatus includes: a first training sample set obtaining unit configured to obtain a first training sample set, where the training sample includes a sample image and is used for characterization Whether the sample image is labeled information of the first category image; the intercepting unit is configured to intercept, for the training samples in the first training sample set, an image of a preset human body part included in the sample image of the training sample to obtain a new sample Image; a sample adding unit configured to add the obtained new sample image and the label information of the new sample image as a new training sample to the first training sample set to obtain a new first training sample set; the training unit, being It is configured to take as input the sample images of the training samples in the new first training sample set, and use the label information corresponding to the input sample images as the desired output, and use the machine learning method to train to obtain the first category image detection model.
  • the sample adding unit is further configured to input the sample image into a pre-trained human body part detection model to obtain detection result information, and the detection result information includes a position of an image of a preset human body part included in the sample image.
  • the human body part detection model is used to characterize the correspondence between the image and the position information of the image of the preset human body part included in the image; the sample image is intercepted based on the obtained position information to obtain a new sample image.
  • the detection result information further includes: category information and confidence of the preset human body part displayed in the image of the preset human body part included in the sample image.
  • the sample adding unit is further configured to: in a descending order of confidence, based on the obtained position information, intercept a preset number of images of a preset human body part from the sample image, and An image of a preset human body part is used as a new sample image.
  • the human body part detection model is obtained by training in the following steps: obtaining a second training sample set, where the training sample includes sample images and label information of the sample images, wherein the label information includes preset human body parts included in the sample image; The position information of the image and the category information of the preset human body part displayed in the sample image; the sample image of the training sample in the second training sample set is used as input, and the label information corresponding to the input sample image is used as the desired output, and the training is Human body part detection model.
  • an embodiment of the present application provides an electronic device.
  • the electronic device includes: one or more processors; a storage device on which one or more programs are stored; Or multiple processors execute, so that the above one or more processors implement the method as described in any implementation manner of the first aspect.
  • an embodiment of the present application provides a computer-readable medium on which a computer program is stored.
  • the program is executed by a processor, the method as described in any implementation manner of the first aspect is implemented.
  • the method and device for generating an image detection model obtained new sample images by intercepting images of preset human body parts included in the sample images. After that, the obtained new sample image and the label information of the new sample image are added as a new training sample to the first training sample set to obtain a new first training sample set. Based on the new first training sample set, a first class image detection model is obtained. Because the new first training sample includes an image of a preset human body part intercepted from the sample image. Therefore, the trained first-type image detection model will detect not only the global information of the image, but also the image of the human body part contained in the image. Therefore, the detection accuracy of the first-type image detection model is improved.
  • FIG. 1 is an exemplary system architecture diagram to which an embodiment of the present application can be applied;
  • FIG. 1 is an exemplary system architecture diagram to which an embodiment of the present application can be applied;
  • FIG. 2 is a flowchart of an embodiment of a method for generating an image detection model according to the present application
  • FIG. 3 is a schematic diagram of an application scenario of a method for generating an image detection model according to the present application
  • FIG. 4 is a flowchart of another embodiment of a method for generating an image detection model according to the present application.
  • FIG. 5 is a schematic diagram of an exemplary training sample in a second training sample set according to the present application, and an exemplary detection result obtained by inputting the training sample into an initial human body part detection model;
  • FIG. 6 is a schematic structural diagram of an embodiment of an apparatus for generating an image detection model according to the present application.
  • FIG. 7 is a flowchart of an embodiment of an image detection method according to the present application.
  • FIG. 8 is a schematic structural diagram of a computer system suitable for implementing an electronic device according to an embodiment of the present application.
  • FIG. 1 illustrates an exemplary system architecture 100 to which a method for generating an image detection model or an apparatus for generating an image detection model according to an embodiment of the present application can be applied.
  • the system architecture 100 may include terminals 101, 102, and 103, a network 104, and a server 105.
  • the network 104 is a medium for providing a communication link between the terminal devices 101, 102, 103 and the server 105.
  • the network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.
  • the terminal devices 101, 102, 103 interact with the server 105 through the network 104, for example, sending the captured images to the server.
  • Various types of camera applications, picture processing applications, etc. can be installed on the terminal devices 101, 102, 103.
  • the terminal devices 101, 102, and 103 may be hardware or software.
  • the terminal devices 101, 102, and 103 may be devices capable of capturing or storing images, including but not limited to: cameras, mobile phones with photographing functions, picture storage servers, and the like.
  • the terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. It can be implemented as multiple software or software modules (for example, to provide photographing services), or it can be implemented as a single software or software module. It is not specifically limited here.
  • the server 105 may be a server that provides various services, such as generating an image detection model based on training samples obtained from the terminal devices 101, 102, and 103.
  • the method for generating an image detection model provided by the embodiment of the present application may be executed by the server 105 or a terminal device.
  • the apparatus for generating the image detection model may be provided in the server 105 or in a terminal device.
  • the image detection model may also be generated in the terminal devices 101, 102, and 103.
  • the method for generating the image detection model may also be executed by the terminal devices 101, 102, and 103.
  • a device for generating an image detection model may also be provided in the terminal devices 101, 102, 103.
  • the exemplary system architecture 100 may be absent from the server 105 and the network 104.
  • the server may be hardware or software.
  • the server can be implemented as a distributed server cluster consisting of multiple servers or as a single server.
  • the server can be implemented as multiple software or software modules (for example, to provide distributed services), or it can be implemented as a single software or software module. It is not specifically limited here.
  • terminal devices, networks, and servers in FIG. 1 are merely exemplary. Depending on the implementation needs, there can be any number of terminal devices, networks, and servers.
  • a flowchart 200 of an embodiment of a method for generating an image detection model according to the present application is shown.
  • the method for generating an image detection model includes the following steps:
  • Step 201 Obtain a first training sample set.
  • the method execution subject generating the image detection model may obtain the first training sample set from the terminal device in a wired connection manner or a wireless connection manner.
  • Each training sample in the first training sample set may include a sample image and annotation information.
  • the label information is used to characterize whether the sample image is a first-class image.
  • the label information may be in various forms.
  • the label information may be a numerical value. For example, "0" indicates that it is not a first category image, and "1" indicates that it is a first category image.
  • the label information may also be text, characters, and so on.
  • the first training sample set may also be stored locally on the execution subject. At this time, the execution body may also directly obtain the first training sample set from the local.
  • the images of the first category may be images of various categories. As an example, it may be a head image, a face image, a bad image, and the like.
  • Step 202 For the training samples in the first training sample set, intercept images of preset human body parts included in the sample images of the training samples to obtain new sample images.
  • each training sample in the first training sample set may intercept the sample image included in the sample image of the training sample in various ways (for example, using various existing screenshot-type applications).
  • each sample image may include an image of at least one preset human body part.
  • the above-mentioned execution subject may intercept images of one or at least two preset human body parts as required to obtain one or at least two new sample images.
  • the preset human body part may be any part of the human body. For example, nose, mouth, and so on. It should be noted that the preset human body part here matches the first category image in step 201.
  • the first category image is a face image.
  • the preset human body parts here may be eyes, mouth, and the like.
  • the first category image is a bad image.
  • the preset human body part here may be a chest, a human reproductive organ, and the like.
  • Step 203 Add the obtained new sample image and the label information of the new sample image as a new training sample to the first training sample set to obtain a new first training sample set.
  • the execution body may add the obtained new sample image and the label information of the new sample image as a new training sample to the first training sample set to obtain a new first training sample set.
  • whether the new sample image is a first-class image can be manually labeled to obtain the label information of the new sample image.
  • one or at least two new sample images may be obtained in step 203.
  • the execution body may add one or at least two new training samples to the first training sample set to obtain a new first training sample set.
  • step 204 a sample image of the training samples in the new first training sample set is used as an input, and label information corresponding to the input sample image is used as a desired output, and a first class image detection model is trained by using a machine learning method.
  • the above-mentioned execution subject may be trained to obtain a first category image detection model based on the new first training sample set.
  • the new first training sample set includes the original training samples, and also includes new training samples.
  • the original training sample and the new training sample are no longer distinguished, and are both called training samples.
  • the above-mentioned execution subject may train the initial first category image detection model based on the new first training sample set to obtain the first category image detection model.
  • the initial first category image detection model may be various image classification networks.
  • the image classification network may be a residual network (Deep Residual Network, ResNet), VGG, or the like.
  • VGG is a classification model proposed by the Visual Geometry Group (VGG) of a university.
  • a sample image of a training sample can be input to an image classification network.
  • an initial value can be set for the image classification network. For example, it could be some different small random numbers.
  • the "small random number” is used to ensure that the network does not enter a saturation state due to excessive weights, which causes training failure. "Different” is used to ensure that the network can learn normally.
  • the detection result of the input sample image can be obtained.
  • the machine classification method is used to train the image classification network. Specifically, the difference between the detection result and the label information calculated by using a preset loss function can be used first.
  • the training end condition here includes but is not limited to at least one of the following: the training time exceeds a preset duration; the number of training times reaches a preset number of times; and the calculated difference is less than a preset difference threshold.
  • FIG. 3 is a schematic diagram of an application scenario of the method for generating an image detection model according to this embodiment.
  • the execution subject of the method for generating an image detection model may be the server 300.
  • the server 300 may first obtain a first training sample set 301.
  • the training sample 302 is a training sample in the first training sample set 301.
  • the training sample 302 includes a sample image 3021 and annotation information 3022.
  • the annotation information 3022 is "1" indicating that the image 3021 is a defective image.
  • the server 300 may sample a breast image included in the present image 3021 to obtain a new sample image 303.
  • the server 300 can perform the same operation on other training samples in the first training sample set 301, and details are not described herein again. After that, the server 300 may add the obtained new sample image 300 and the acquired labeling information “1” marked by a technician to the first training sample set 301 as new training samples. Similarly, other new training samples obtained may also be added to the first training sample set 301. After that, a new first training sample set 301 'can be obtained.
  • the server 300 takes as input the sample images of the training samples in the new first training sample set 301 ′, and uses the labeling information corresponding to the input sample images as the desired output.
  • the machine learning method is used to classify the network. Training is performed to obtain a bad image detection model.
  • the method provided by the foregoing embodiments of the present application obtains a new sample image by sampling an image of a preset human body part included in the image. After that, the obtained new sample image and the label information of the new sample image are added as a new training sample to the first training sample set to obtain a new first training sample set. Based on the new first training sample set, a first class image detection model is obtained. Because the new first training sample includes an image of a preset human body part intercepted from the sample image. Therefore, the trained first-type image detection model will detect not only the global information of the image, but also the image of the human body part contained in the image. Therefore, the detection accuracy of the first-type image detection model is improved.
  • FIG. 4 illustrates a flowchart 400 of still another embodiment of a method for generating an image detection model.
  • the process 400 of the method for generating an image detection model includes the following steps:
  • Step 401 Obtain a first training sample set.
  • step 401 the specific implementation of step 401 and the technical effects it brings are similar to step 201 in the embodiment corresponding to FIG. 2, and details are not described herein again.
  • Step 402 For the training samples in the first training sample set, perform the following training steps:
  • Step 4021 Input the sample image into a pre-trained human body part detection model to obtain position information of an image of a preset human body part included in the sample image, category information of the preset human body part displayed in the image of the preset human body part, and Confidence.
  • the confidence degree is used to indicate the credibility of the category information.
  • confidence can be expressed as a probability value.
  • the human body part detection model is used to represent the image and the position information of the image of the preset human body part included in the image, the correspondence between the category information of the preset human body part and the confidence displayed in the image of the preset human body part relationship.
  • the human body part detection model is obtained by training in the following steps:
  • the executing body can obtain a second training sample set.
  • the second training sample set here is for distinguishing from the first training sample set above.
  • the first and second are not limitations on the training sample set.
  • Each training sample in the second training sample set may include a sample image and label information of the sample image.
  • the annotation information includes position information of an image of a preset human body part included in the sample image and category information of the preset human body part displayed in the sample image.
  • the position information is used to represent a position of an image of a preset human body part relative to a sample image. Position information comes in various forms, such as callout boxes, coordinates, and so on.
  • the category information of the preset human body part is used to indicate the category of the preset human body part.
  • the preset human body part may be at least one human body part.
  • the preset human body part may include three parts of a female chest, a male genitalia, and a female genitalia.
  • the category information of the preset human body part may be “00”, “01”, and “10”, which are respectively used to indicate these three parts.
  • FIG. 5 illustrates an exemplary training sample 500 in the second training sample set.
  • the training sample includes a sample image 501, a labeling frame 502, and category information "01".
  • the executing body may take the sample image of the training samples in the second training sample set as input, and use the label information corresponding to the input sample image as the desired output to train and obtain a human part detection model.
  • the sample images of the training samples in the second training sample set may be input into the initial human body part detection model.
  • the initial human body part detection model may be various target detection networks. As an example, it can be an existing SSD (Single Shot MultiBox Detector) or YOLO (You Only Look Out).
  • an initial value can be set for the initial human body part detection model. For example, it could be some different small random numbers. The "small random number" is used to ensure that the network does not enter a saturation state due to excessive weights, which causes training failure.
  • the detection result of the input sample image can be obtained.
  • the figure shows the detection results obtained by inputting the sample image 501 into the initial human body part detection model. It can be seen that the detection result includes a label box 502 ', category information "01”, and a confidence level of 0.92. Among them, the confidence level of 0.92 can indicate that the probability of displaying the male genitals in the sample image 501 is 92%.
  • the machine learning method is used to train the initial human body part detection model.
  • the difference between the detection result and the label information calculated by using a preset loss function can be used first. Then, the parameters of the initial human body part detection model can be adjusted based on the obtained differences, and if the preset training end condition is met, the training is ended, and the trained initial human body part detection model is used as the human body part detection model.
  • the training end condition here includes but is not limited to at least one of the following: the training time exceeds a preset duration; the number of training times reaches a preset number of times; and the calculated difference is less than a preset difference threshold.
  • BP Back Propagation, Back Propagation
  • SGD Spochastic Gradient Descent, Stochastic Gradient Descent
  • the execution subject of the training step and the method of generating the image detection model may be the same or different. If they are the same, the executing subject can store the network structure and parameter values of the human body part detection model locally after training to obtain the human body part detection model. If they are different, after the execution subject of the training step obtains the human body part detection model, the network structure and parameter values of the model may be sent to the execution subject of the method for generating an image detection model.
  • step 4022 according to the order of confidence from the largest to the smallest, based on the obtained position information, a predetermined number of images of the preset human body parts are intercepted from the sample image, and The image serves as the new sample image.
  • one sample image may include an image of at least one preset human body part. Therefore, the execution subject can intercept a preset number of images of a preset human body part from the sample image based on the obtained position information in descending order of confidence, and use the intercepted image of the preset human body part as a new Sample image.
  • Step 403 Add the obtained new sample image and the label information of the new sample image as a new training sample to the first training sample set to obtain a new first training sample set.
  • step 404 a sample image of the training samples in the new first training sample set is used as input, and label information corresponding to the input sample image is used as a desired output, and a first class image detection model is trained by using a machine learning method.
  • steps 403 and 404 for the specific processing of steps 403 and 404 and the technical effects brought by them, reference may be made to steps 203 and 204 in the embodiment corresponding to FIG. 2, and details are not described herein again.
  • the method for generating an image detection model in this embodiment uses a human body part detection model to obtain a preset number of preset human body parts included in a sample image. image. Compared with manual labeling, the labeling efficiency can be improved.
  • this application provides an embodiment of an apparatus for generating an image detection model.
  • the apparatus embodiment corresponds to the method embodiment shown in FIG. 2.
  • the device can be specifically applied to various electronic devices.
  • the apparatus 600 for generating an image detection model in this embodiment includes a first training sample set acquiring unit 601, a cutting unit 602, a sample adding unit 603, and a training unit 604.
  • the first training sample set obtaining unit 601 is configured to obtain a first training sample set
  • the training samples include a sample image and label information used to characterize whether the sample image is a first class image.
  • the interception unit 602 is configured to, for a training sample in the first training sample set, intercept an image of a preset human body part included in a sample image of the training sample to obtain a new sample image.
  • the sample adding unit 603 is configured to add the obtained new sample image and the label information of the new sample image as a new training sample to the first training sample set to obtain a new first training sample set.
  • the training unit 604 is configured to take as input the sample images of the training samples in the new first training sample set, and use the label information corresponding to the input sample images as the desired output, and use the machine learning method to train to obtain the first category image Detection model.
  • the specific implementation of the first training sample set acquisition unit 601, the interception unit 602, the sample addition unit 603, and the training unit 604 and the technical effects brought by the apparatus 600 for generating an image detection model in this embodiment may refer to FIG. 2 Steps 201-204 of the embodiment are not repeated here.
  • the sample adding unit 603 may be further configured to: input the sample image into a pre-trained human body part detection model to obtain detection result information.
  • the detection result information includes position information of an image of a preset human body part included in the sample image.
  • the human body part detection model is used to characterize the correspondence between the image and the position information of the preset human body part image included in the image. And intercepting the sample image based on the obtained position information to obtain a new sample image.
  • the detection result information further includes: category information and confidence of the preset human body part displayed in the image of the preset human body part included in the sample image.
  • the sample adding unit 603 may be further configured to: in a descending order of confidence, based on the obtained position information, intercept a preset number of pre-preparations from the sample image. Set the image of the human body part and the captured image of the preset human body part as the new sample image.
  • the human part detection model is obtained by training in the following steps: obtaining a second training sample set, where the training sample includes sample images and label information of the sample images, where the label information includes the sample images The position information of the image of the preset human body part and the category information of the preset human body part included in the sample image; the sample image of the training sample in the second training sample set is used as an input, and the annotation corresponding to the input sample image is labeled The information is output as expected, and a human body part detection model is trained.
  • the above-mentioned intercepting unit 602 may obtain a new sample image by intercepting the image of a preset human body part included in the image.
  • the sample adding unit 603 adds the obtained new sample image and the label information of the new sample image as a new training sample to the first training sample set to obtain a new first training sample set.
  • the training unit 604 may train a first category image detection model based on the new first training sample set. Because the new first training sample includes an image of a preset human body part intercepted from the sample image. Therefore, the trained first-type image detection model will detect not only the global information of the image, but also the image of the human body part contained in the image. Therefore, the detection accuracy of the first-type image detection model is improved.
  • the process 700 of the image detection method includes the following steps:
  • Step 701 Acquire an image to be detected.
  • the image to be detected may be any image.
  • the determination of the image to be detected can be specified by a technician, or it can be filtered according to certain conditions.
  • Step 702 Input an image to be detected into a first category image detection model, and obtain detection result information used to characterize whether the image to be detected is a first category image.
  • the first-type image detection model is generated according to the methods described in various implementation manners of the embodiment corresponding to FIG. 2.
  • FIG. 8 illustrates a schematic structural diagram of a computer system 800 suitable for implementing a server according to an embodiment of the present application.
  • the server shown in FIG. 8 is merely an example, and should not impose any limitation on the functions and scope of use of the embodiments of the present application.
  • the computer system 800 includes a central processing unit (CPU) 801, which can be loaded into a random access memory (RAM) 803 according to a program stored in a read-only memory (ROM) 802 or from a storage section 808. Instead, perform various appropriate actions and processes.
  • RAM random access memory
  • ROM read-only memory
  • various programs and data required for the operation of the system 800 are also stored.
  • the CPU 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804.
  • An input / output (I / O) interface 805 is also connected to the bus 804.
  • the following components are connected to the I / O interface 805: an input portion 806 including a keyboard, a mouse, etc .; an output portion 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc .; and a speaker; a storage portion 808 including a hard disk, etc. ; And a communication section 809 including a network interface card such as a LAN card, a modem, and the like. The communication section 809 performs communication processing via a network such as the Internet.
  • the driver 810 is also connected to the I / O interface 805 as needed.
  • a removable medium 811 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed, so that a computer program read therefrom is installed into the storage section 808 as needed.
  • the process described above with reference to the flowchart may be implemented as a computer software program.
  • embodiments of the present disclosure include a computer program product including a computer program carried on a computer-readable medium, the computer program containing program code for performing a method shown in a flowchart.
  • the computer program may be downloaded and installed from a network through the communication section 809, and / or installed from a removable medium 811.
  • CPU central processing unit
  • the computer-readable medium described in this application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two.
  • the computer-readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable Programming read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the foregoing.
  • a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in combination with an instruction execution system, apparatus, or device.
  • a computer-readable signal medium may include a data signal that is included in baseband or propagated as part of a carrier wave, and which carries computer-readable program code. Such a propagated data signal may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.
  • the computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, and the computer-readable medium may send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device .
  • Program code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
  • Computer program code for performing the operations of this application may be written in one or more programming languages, or a combination thereof, including programming languages such as Java, Smalltalk, C ++, and also conventional Procedural programming language—such as "C" or a similar programming language.
  • the program code can be executed entirely on the user's computer, partly on the user's computer, as an independent software package, partly on the user's computer, partly on a remote computer, or entirely on a remote computer or server.
  • 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 (such as through an Internet service provider) Internet connection).
  • LAN local area network
  • WAN wide area network
  • Internet service provider Internet service provider
  • each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, which contains one or more functions to implement a specified logical function Executable instructions.
  • the functions labeled in the blocks may also occur in a different order than those labeled in the drawings. For example, two blocks represented one after the other may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved.
  • each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts can be implemented by a dedicated hardware-based system that performs the specified function or operation , Or it can be implemented with a combination of dedicated hardware and computer instructions.
  • the units described in the embodiments of the present application may be implemented by software or hardware.
  • the described unit may also be provided in a processor, for example, it may be described as: a processor includes a first training sample set acquisition unit, an interception unit, a sample addition unit, and a training unit.
  • a processor includes a first training sample set acquisition unit, an interception unit, a sample addition unit, and a training unit.
  • the name of these units does not constitute a limitation on the unit itself in some cases.
  • the first training sample set acquisition unit may also be described as a “unit that acquires the first training sample set”.
  • the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or may exist separately without being assembled into the electronic device in.
  • the computer-readable medium carries one or more programs.
  • the electronic device is configured to obtain a first training sample set, where the training samples include a sample image and a sample image for characterizing the sample image.
  • the new samples will be obtained
  • the label information of the image and the new sample image is added as a new training sample to the first training sample set to obtain a new first training sample set. Taking the sample image of the training samples in the new first training sample set as input, the The labeled information corresponding to the input sample image is used as the desired output, and a first-class image detection model is trained by using a machine learning method.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

本申请实施例公开了生成图像检测模型的方法和装置。该方法的一具体实施方式包括：获取第一训练样本集合；对于第一训练样本集合中的训练样本，截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像；将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合；基于新的第一训练样本集合，利用机器学习的方法，训练得到第一类别图像检测模型。该实施方式实现了提高第一类别图像检测模型的检测准确率。

Description

生成图像检测模型的方法和装置
本专利申请要求于2018年7月6日提交的、申请号为201810734679.5、申请人为北京字节跳动网络技术有限公司、发明名称为“生成图像检测模型的方法和装置”的中国专利申请的优先权，该申请的全文以引用的方式并入本申请中。
技术领域
本申请实施例涉及计算机技术领域，具体涉及生成图像检测模型的方法和装置。
背景技术
随着互联网的快速发展，尤其是移动互联网的普及，各种内容的视频或图像层出不穷。目前，主要采取人工审核的方式对这些视频或图像的内容进行审核。
发明内容
本申请实施例提出了生成图像检测模型的方法和装置。
第一方面，本申请实施例提供了一种生成图像检测模型的方法，该方法包括：获取第一训练样本集合，训练样本包括样本图像和用于表征样本图像是否为第一类别图像的标注信息；对于第一训练样本集合中的训练样本，截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像；将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合；将新的第一训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，利用机器学习的方法，训练得到第一类别图像检测模型。
在一些实施例中，截取该训练样本的样本图像中包括的预设人体 部位的图像，得到新的样本图像，包括：将该样本图像输入预先训练的人体部位检测模型，得到检测结果信息，检测结果信息包括该样本图像中包括的预设人体部位的图像的位置信息，其中，人体部位检测模型用于表征图像和图像中包括的预设人体部位的图像的位置信息的对应关系；基于得到的位置信息对该样本图像进行截取，得到新的样本图像。
在一些实施例中，检测结果信息还包括：该样本图像中包括的预设人体部位的图像中显示的预设人体部位的类别信息和置信度。
在一些实施例中，基于得到的位置信息对该样本图像进行截取，得到新的样本图像，包括：按照置信度由大到小的顺序，基于得到的位置信息，从该样本图像中截取预设数量的预设人体部位的图像，以及将截取的预设人体部位的图像作为新的样本图像。
在一些实施例中，人体部位检测模型通过以下步骤训练得到：获取第二训练样本集合，训练样本包括样本图像和样本图像的标注信息，其中，标注信息包括样本图像中包含的预设人体部位的图像的位置信息和样本图像中显示的预设人体部位的类别信息；将第二训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，训练得到人体部位检测模型。
第二方面，本申请实施例提供了一种图像检测方法，该方法包括：获取待检测图像；将待检测图像输入第一类别图像检测模型，得到用于表征待检测图像是否为第一类别图像的检测结果信息，其中，第一类别图像检测模型是按照如第一方面中任一实现方式描述的方法生成的。
第三方面，本申请实施例提供了一种生成图像检测模型的装置，该装置包括：第一训练样本集合获取单元，被配置成获取第一训练样本集合，训练样本包括样本图像和用于表征样本图像是否为第一类别图像的标注信息；截取单元，被配置成对于第一训练样本集合中的训练样本，截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像；样本添加单元，被配置成将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集 合，得到新的第一训练样本集合；训练单元，被配置成将新的第一训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，利用机器学习的方法，训练得到第一类别图像检测模型。
在一些实施例中，样本添加单元进一步被配置成：将该样本图像输入预先训练的人体部位检测模型，得到检测结果信息，检测结果信息包括该样本图像中包括的预设人体部位的图像的位置信息，其中，人体部位检测模型用于表征图像和图像中包括的预设人体部位的图像的位置信息的对应关系；基于得到的位置信息对该样本图像进行截取，得到新的样本图像。
在一些实施例中，检测结果信息还包括：该样本图像中包括的预设人体部位的图像中显示的预设人体部位的类别信息和置信度。
在一些实施例中，样本添加单元进一步被配置成：按照置信度由大到小的顺序，基于得到的位置信息，从该样本图像中截取预设数量的预设人体部位的图像，以及将截取的预设人体部位的图像作为新的样本图像。
在一些实施例中，人体部位检测模型通过以下步骤训练得到：获取第二训练样本集合，训练样本包括样本图像和样本图像的标注信息，其中，标注信息包括样本图像中包含的预设人体部位的图像的位置信息和样本图像中显示的预设人体部位的类别信息；将第二训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，训练得到人体部位检测模型。
第四方面，本申请实施例提供了一种电子设备，该电子设备包括：一个或多个处理器；存储装置，其上存储有一个或多个程序；当上述一个或多个程序被上述一个或多个处理器执行，使得上述一个或多个处理器实现如第一方面中任一实现方式描述的方法。
第五方面，本申请实施例提供了一种计算机可读介质，其上存储有计算机程序，上述程序被处理器执行时实现如第一方面中任一实现方式描述的方法。
本申请实施例提供的生成图像检测模型的方法和装置，通过截取 样本图像中包括的预设人体部位的图像，得到新的样本图像。之后，将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合。并基于新的第一训练样本集合训练得到第一类别图像检测模型。由于新的第一训练样本中包含了从样本图像中截取的预设人体部位的图像。因此，训练完成的第一类别图像检测模型不仅会针对图像的全局信息进行检测，而且会针对图像中包含的人体部位图像的进行检测。从而提高了第一类别图像检测模型的检测准确率。
附图说明
通过阅读参照以下附图所作的对非限制性实施例所作的详细描述，本申请的其它特征、目的和优点将会变得更明显：
图1是本申请的一个实施例可以应用于其中的示例性系统架构图；
图2是根据本申请的生成图像检测模型的方法的一个实施例的流程图；
图3是根据本申请的生成图像检测模型的方法的一个应用场景的示意图；
图4是根据本申请的生成图像检测模型的方法的又一个实施例的流程图；
图5是根据本申请的第二训练样本集合中的一个示例性训练样本，以及将训练样本输入初始人体部位检测模型后得到的示例性检测结果的示意图；
图6根据本申请的生成图像检测模型的装置的一个实施例的结构示意图；
图7是根据本申请的图像检测方法的一个实施例的流程图；
图8是适于用来实现本申请实施例的电子设备的计算机系统的结构示意图。
具体实施方式
下面结合附图和实施例对本申请作进一步的详细说明。可以理解的是，此处所描述的具体实施例仅仅用于解释相关发明，而非对该发明的限定。另外还需要说明的是，为了便于描述，附图中仅示出了与有关发明相关的部分。
需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。下面将参考附图并结合实施例来详细说明本申请。
图1示出了可以应用本申请实施例的生成图像检测模型的方法或生成图像检测模型的装置的示例性系统架构100。
如图1所示，系统架构100可以包括终端101、102、103，网络104和服务器105。网络104用以在终端设备101、102、103和服务器105之间提供通信链路的介质。网络104可以包括各种连接类型，例如有线、无线通信链路或者光纤电缆等等。
终端设备101、102、103通过网络104与服务器105交互，例如将拍摄的图像发送至服务器。终端设备101、102、103上可以安装有各类拍照应用、图片处理应用等。
终端设备101、102、103可以是硬件，也可以是软件。当终端设备101、102、103为硬件时，可以是能够拍摄或者存储图像的设备，包括但不限于：照相机、具备拍照功能的手机、图片存储服务器等等。当终端设备101、102、103为软件时，可以安装在上述所列举的电子设备中。其可以实现成多个软件或软件模块(例如用来提供拍照服务)，也可以实现成单个软件或软件模块。在此不做具体限定。
服务器105可以是提供各种服务的服务器，例如基于从终端设备101、102、103获取的训练样本生成图像检测模型。
需要说明的是，本申请实施例所提供的生成图像检测模型的方法可以由服务器105执行，也可以由终端设备执行。相应地，生成图像检测模型的装置可以设置于服务器105中，也可以设置于终端设备中。
需要说明的是，终端设备101、102、103中也可以生成图像检测模型。此时，生成图像检测模型的方法也可以由终端设备101、102、103执行。相应地，生成图像检测模型的装置也可以设置于终端设备 101、102、103中。此时，示例性系统架构100可以不存在服务器105和网络104。
需要说明的是，服务器可以是硬件，也可以是软件。当服务器为硬件时，可以实现成多个服务器组成的分布式服务器集群，也可以实现成单个服务器。当服务器为软件时，可以实现成多个软件或软件模块(例如用来提供分布式服务)，也可以实现成单个软件或软件模块。在此不做具体限定。
应该理解，图1中的终端设备、网络和服务器的数目仅仅是示意性的。根据实现需要，可以具有任意数目的终端设备、网络和服务器。
继续参考图2，示出了根据本申请的生成图像检测模型的方法的一个实施例的流程200。该生成图像检测模型的方法，包括以下步骤：
步骤201，获取第一训练样本集合。
在本实施例中，生成图像检测模型的方法执行主体可以通过有线连接方式或者无线连接方式从终端设备获取第一训练样本集合。其中，第一训练样本集合中的每个训练样本可以包括样本图像和标注信息。标注信息用于表征样本图像是否为第一类别图像。这里，标注信息可以是各种形式。作为示例，标注信息可以是数值。例如，用“0”表示不是第一类别图像，用“1”表示是第一类别图像。作为示例，标注信息还可以是文字、字符等等。此外，上述第一训练样本集合也可以存储于执行主体本地。此时，上述执行主体也可以从本地直接获取第一训练样本集合。
在本实施例中，第一类别图像可以是各种类别的图像。作为示例，可以是头部图像、面部图像、不良图像等等。
步骤202，对于第一训练样本集合中的训练样本，截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像。
在本实施例中，对于第一训练样本集合中的每个训练样本，上述执行主体可以通过各种方式(例如利用现有的各种截图类应用)，截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像。实践中，每个样本图像中可以包括至少一个预设人体部位的图像。此时，上述执行主体可以根据需要截取一个或至少两个预设人体 部位的图像，得到一个或至少两个新的样本图像。其中，预设人体部位可以是人体的任意部位。例如，鼻子、嘴巴等等。需要说明的是，此处的预设人体部位与步骤201中的第一类别图像是相匹配的。例如，第一类别图像是面部图像。那么，此处的预设人体部位可以是眼睛、嘴巴等。又如，第一类别图像是不良图像。那么，此处的预设人体部位可以是胸部、人体生殖器官等等。
步骤203，将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合。
在本实施例中，上述执行主体可以将得到的新的样本图像与新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合。作为示例，可以通过人工对新的样本图像是否为第一类别图像进行标注，从而得到新的样本图像的标注信息。需要说明的是，步骤203中可以得到一个或至少两个新的样本图像。相应的，上述执行主体可以将一个或至少两个新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合。
步骤204，将新的第一训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，利用机器学习的方法，训练得到第一类别图像检测模型。
在本实施例中，上述执行主体可以基于新的第一训练样本集合，训练得到第一类别图像检测模型。其中，新的第一训练样本集合中包括原来的训练样本，还包括新的训练样本。在基于新的第一训练样本集合进行训练时，不再区分原来的训练样本和新的训练样本，均称为训练样本。
本实施例中，上述执行主体可以基于新的第一训练样本集合，对初始第一类别图像检测模型进行训练，得到第一类别图像检测模型。其中，初始第一类别图像检测模型可以是各种图像分类网络。作为示例，图像分类网络可以是残差网络(Deep Residual Network，ResNet)、VGG等等。VGG是某大学的视觉几何小组(Visual Geometry Group，VGG)提出的分类模型。
具体来说，可以将训练样本的样本图像输入图像分类网络。实践中，可以为图像分类网络设置初始值。例如，可以是一些不同的小随机数。“小随机数”用来保证网络不会因权值过大而进入饱和状态，从而导致训练失败，“不同”用来保证网络可以正常地学习。之后，可以得到输入的样本图像的检测结果。以与输入的样本图像对应的标注信息作为图像分类网络的期望输出，利用机器学习方法训练图像分类网络。具体来说可以首先利用预设的损失函数计算得到的检测结果与标注信息之间的差异。然后，可以基于所得到的差异，调整图像分类网络的参数，并在满足预设的训练结束条件的情况下，结束训练，并将训练后的图像分类网络作为第一类别图像检测模型。这里的训练结束条件包括但不限于以下至少一项：训练时间超过预设时长；训练次数达到预设次数；计算所得的差异小于预设差异阈值。
继续参见图3，图3是根据本实施例的生成图像检测模型的方法的应用场景的一个示意图。在图3的应用场景中，生成图像检测模型的方法的执行主体可以是服务器300。服务器300可以首先获取第一训练样本集合301。其中，训练样本302为第一训练样本集合301中的一个训练样本。训练样本302包括样本图像3021和标注信息3022。标注信息3022为“1”表示图像3021为不良图像。服务器300可以截取样本图像3021中包括的胸部的图像，得到新的样本图像303。可以理解，服务器300对于第一训练样本集合301中的其他训练样本可以执行同样的操作，在此不再赘述。之后，服务器300可以将得到的新的样本图像300和获取的由技术人员标注的标注信息“1”作为新的训练样本添加到第一训练样本集合301。同样的，对于得到的其他的新的训练样本也可以添加到第一训练样本集合301。之后，可以得到新的第一训练样本集合301'。
在此基础上，服务器300将新的第一训练样本集合301'中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，利用机器学习的方法，对分类网络进行训练，训练得到不良图像检测模型。
本申请的上述实施例提供的方法通过截取样本图像中包括的预设 人体部位的图像，得到新的样本图像。之后，将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合。并基于新的第一训练样本集合训练得到第一类别图像检测模型。由于新的第一训练样本中包含了从样本图像中截取的预设人体部位的图像。因此，训练完成的第一类别图像检测模型不仅会针对图像的全局信息进行检测，而且会针对图像中包含的人体部位图像的进行检测。从而提高了第一类别图像检测模型的检测准确率。
进一步参考图4，其示出了生成图像检测模型的方法又一个实施例的流程400。该生成图像检测模型的方法的流程400，包括以下步骤：
步骤401，获取第一训练样本集合。
本实施例中，步骤401的具体实现以及所带来的技术效果与图2对应的实施例中的步骤201类似，在此不再赘述。
步骤402，对于第一训练样本集合中的训练样本，执行如下训练步骤：
步骤4021，将该样本图像输入预先训练的人体部位检测模型，得到该样本图像中包括的预设人体部位的图像的位置信息、预设人体部位的图像中显示的预设人体部位的类别信息和置信度。其中，置信度用于表示类别信息的可信程度。实践中，置信度可以采用概率值表示。
在本实施例中，人体部位检测模型用于表征图像和图像中包括的预设人体部位的图像的位置信息、预设人体部位的图像中显示的预设人体部位的类别信息和置信度的对应关系。
在本实施例的一些可选的实现方式中，人体部位检测模型通过以下步骤训练得到：
第一步，执行主体可以获取第二训练样本集合。可以理解，这里的第二训练样本集合是为了与上文中的第一训练样本集合区分。其中，第一、第二并不是对于训练样本集合的限定。第二训练样本集合中的每个训练样本可以包括样本图像和样本图像的标注信息。其中，标注信息包括样本图像中包含的预设人体部位的图像的位置信息和样本图 像中显示的预设人体部位的类别信息。其中，位置信息用于表征预设人体部位的图像相对于样本图像的位置。位置信息有各种形式，例如，标注框、坐标等等。预设人体部位的类别信息用于表示预设人体部位的类别。其中，预设人体部位可以是至少一个人体部位。作为示例，预设人体部位可以包括女性胸部、男性生殖器、女性生殖器这三个部位。那么，预设人体部位的类别信息可以是“00”、“01”、“10”，分别用于表示这三个部位。这里可以参考图5，图中示出了第二训练样本集合中的一个示例性训练样本500。其中，训练样本包括样本图像501、标注框502和类别信息“01”。
第二步，执行主体可以将第二训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，训练得到人体部位检测模型。具体来说，可以将第二训练样本集合中的训练样本的样本图像输入初始人体部位检测模型。其中，初始人体部位检测模型可以是各种目标检测网络。作为示例，可以是现有的SSD(Single Shot MultiBox Detector)或YOLO(You Only Look Once)等等。实践中，可以为初始人体部位检测模型设置初始值。例如，可以是一些不同的小随机数。“小随机数”用来保证网络不会因权值过大而进入饱和状态，从而导致训练失败，“不同”用来保证网络可以正常地学习。之后，可以得到输入的样本图像的检测结果。继续参考图5，图中示出了将样本图像501输入初始人体部位检测模型，得到的检测结果。可以看到，检测结果包括标注框502'、类别信息“01”和置信度0.92。其中，置信度0.92可以表示样本图像501中显示男性生殖器的概率为92％。以与输入的样本图像的标注信息作为初始人体部位检测模型的期望输出，利用机器学习方法训练初始人体部位检测模型。具体来说可以首先利用预设的损失函数计算得到的检测结果与标注信息之间的差异。然后，可以基于所得到的差异，调整初始人体部位检测模型的参数，并在满足预设的训练结束条件的情况下，结束训练，并将训练后的初始人体部位检测模型作为人体部位检测模型。这里的训练结束条件包括但不限于以下至少一项：训练时间超过预设时长；训练次数达到预设次数；计算所得的差异小于预设差异阈值。
这里可以采用各种方式基于所得到的检测结果与输入的训练样本对应的标注信息之间的差异，调整初始人体部位检测模型的参数。例如，可以采用BP(Back Propagation，反向传播)算法或者SGD(Stochastic Gradient Descent，随机梯度下降)算法来调整初始图像分类网络的参数。
需要说明的是，训练步骤的执行主体与生成图像检测模型的方法的执行主体可以相同，也可以不同。若相同，执行主体可以在训练得到人体部位检测模型后，将人体部位检测模型的网络结构和参数值存储于本地。若不同，训练步骤的执行主体在训练得到人体部位检测模型后，可以将模型的网络结构和参数值发送至生成图像检测模型的方法的执行主体。
返回参考图4，步骤4022，按照置信度由大到小的顺序，基于得到的位置信息，从该样本图像中截取预设数量的预设人体部位的图像，以及将截取的预设人体部位的图像作为新的样本图像。
在本实施例中，由于一个样本图像中可以包括至少一个预设人体部位的图像。因此，执行主体可以按照置信度由大到小的顺序，基于得到的位置信息，从该样本图像中截取预设数量的预设人体部位的图像，以及将截取的预设人体部位的图像作为新的样本图像。
步骤403，将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合。
步骤404，将新的第一训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，利用机器学习的方法，训练得到第一类别图像检测模型。
本实施例中，步骤403、404的具体处理及其所带来的技术效果可以参考图2对应的实施例中的步骤203、204，在此不再赘述。
从图4中可以看出，与图2对应的实施例相比，本实施例中的生成图像检测模型的方法利用人体部位检测模型，得到样本图像中包括的预设数量的预设人体部位的图像。与人工标注相比，可以提高标注效率。
进一步参考图6，作为对上述各图所示方法的实现，本申请提供了一种生成图像检测模型的装置的一个实施例，该装置实施例与图2所示的方法实施例相对应，该装置具体可以应用于各种电子设备中。
如图6所示，本实施例的生成图像检测模型的装置600包括：第一训练样本集合获取单元601、截取单元602、样本添加单元603和训练单元604。其中，第一训练样本集合获取单元601被配置成获取第一训练样本集合，训练样本包括样本图像和用于表征样本图像是否为第一类别图像的标注信息。截取单元602被配置成对于第一训练样本集合中的训练样本，截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像。样本添加单元603被配置成将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合。训练单元604被配置成将新的第一训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，利用机器学习的方法，训练得到第一类别图像检测模型。
本实施例中的生成图像检测模型的装置600包括的第一训练样本集合获取单元601、截取单元602、样本添加单元603和训练单元604的具体实现以及所带来的技术效果可以参考图2对应的实施例的步骤201-204，在此不再赘述。
在本实施例的一些可选的实现方式中，样本添加单元603可以进一步被配置成：将该样本图像输入预先训练的人体部位检测模型，得到检测结果信息。检测结果信息包括该样本图像中包括的预设人体部位的图像的位置信息。其中，人体部位检测模型用于表征图像和图像中包括的预设人体部位的图像的位置信息的对应关系。以及基于得到的位置信息对该样本图像进行截取，得到新的样本图像。
在本实施例的一些可选的实现方式中，检测结果信息还包括：该样本图像中包括的预设人体部位的图像中显示的预设人体部位的类别信息和置信度。
在本实施例的一些可选的实现方式中，样本添加单元603可以进 一步被配置成：按照置信度由大到小的顺序，基于得到的位置信息，从该样本图像中截取预设数量的预设人体部位的图像，以及将截取的预设人体部位的图像作为新的样本图像。
在本实施例的一些可选的实现方式中，人体部位检测模型通过以下步骤训练得到：获取第二训练样本集合，训练样本包括样本图像和样本图像的标注信息，其中，标注信息包括样本图像中包含的预设人体部位的图像的位置信息和样本图像中包含的预设人体部位的类别信息；将第二训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，训练得到人体部位检测模型。
在本实施例中，对于第一训练样本集合获取单元601获取的第一训练样本集合，上述截取单元602可以通过截取样本图像中包括的预设人体部位的图像，得到新的样本图像。之后，样本添加单元603将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合。训练单元604可以基于新的第一训练样本集合训练得到第一类别图像检测模型。由于新的第一训练样本中包含了从样本图像中截取的预设人体部位的图像。因此，训练完成的第一类别图像检测模型不仅会针对图像的全局信息进行检测，而且会针对图像中包含的人体部位图像的进行检测。从而提高了第一类别图像检测模型的检测准确率。
继续参考图7，其示出了图像检测方法的一个实施例的流程700。该图像检测方法的流程700包括以下步骤：
步骤701，获取待检测图像。其中，待检测图像可以是任意图像。待检测图像的确定可以由技术人员指定，也可以根据一定的条件筛选。
步骤702，将待检测图像输入第一类别图像检测模型，得到用于表征待检测图像是否为第一类别图像的检测结果信息。
本实施例中，第一类别图像检测模型是按照如图2对应的实施例的各种实现方式描述的方法生成的。
下面参考图8，其示出了适于用来实现本申请实施例的服务器的计算机系统800的结构示意图。图8示出的服务器仅仅是一个示例，不应对本申请实施例的功能和使用范围带来任何限制。
如图8所示，计算机系统800包括中央处理单元(CPU)801，其可以根据存储在只读存储器(ROM)802中的程序或者从存储部分808加载到随机访问存储器(RAM)803中的程序而执行各种适当的动作和处理。在RAM 803中，还存储有系统800操作所需的各种程序和数据。CPU 801、ROM 802以及RAM 803通过总线804彼此相连。输入/输出(I/O)接口805也连接至总线804。
以下部件连接至I/O接口805：包括键盘、鼠标等的输入部分806；包括诸如阴极射线管(CRT)、液晶显示器(LCD)等以及扬声器等的输出部分807；包括硬盘等的存储部分808；以及包括诸如LAN卡、调制解调器等的网络接口卡的通信部分809。通信部分809经由诸如因特网的网络执行通信处理。驱动器810也根据需要连接至I/O接口805。可拆卸介质811，诸如磁盘、光盘、磁光盘、半导体存储器等等，根据需要安装在驱动器810上，以便于从其上读出的计算机程序根据需要被安装入存储部分808。
特别地，根据本公开的实施例，上文参考流程图描述的过程可以被实现为计算机软件程序。例如，本公开的实施例包括一种计算机程序产品，其包括承载在计算机可读介质上的计算机程序，该计算机程序包含用于执行流程图所示的方法的程序代码。在这样的实施例中，该计算机程序可以通过通信部分809从网络上被下载和安装，和/或从可拆卸介质811被安装。在该计算机程序被中央处理单元(CPU)801执行时，执行本申请的方法中限定的上述功能。
需要说明的是，本申请所述的计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质或者是上述两者的任意组合。计算机可读存储介质例如可以是——但不限于——电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子可以包括但不限于：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机访问存储器(RAM)、 只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本申请中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。而在本申请中，计算机可读的信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。计算机可读的信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。计算机可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于：无线、电线、光缆、RF等等，或者上述的任意合适的组合。
可以以一种或多种程序设计语言或其组合来编写用于执行本申请的操作的计算机程序代码，所述程序设计语言包括面向对象的程序设计语言—诸如Java、Smalltalk、C++，还包括常规的过程式程序设计语言—诸如”C”语言或类似的程序设计语言。程序代码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)。
附图中的流程图和框图，图示了按照本申请各种实施例的系统、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、或代码的一部分，该模块、程序段、或代码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。也应当注意，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，两个接连地表示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操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。
描述于本申请实施例中所涉及到的单元可以通过软件的方式实现，也可以通过硬件的方式来实现。所描述的单元也可以设置在处理器中，例如，可以描述为：一种处理器包括第一训练样本集合获取单元、截取单元、样本添加单元和训练单元。其中，这些单元的名称在某种情况下并不构成对该单元本身的限定，例如，第一训练样本集合获取单元还可以被描述为“获取第一训练样本集合的单元”。
作为另一方面，本申请还提供了一种计算机可读介质，该计算机可读介质可以是上述实施例中描述的电子设备中所包含的；也可以是单独存在，而未装配入该电子设备中。上述计算机可读介质承载有一个或者多个程序，当上述一个或者多个程序被该电子设备执行时，使得该电子设备：获取第一训练样本集合，训练样本包括样本图像和用于表征样本图像是否为第一类别图像的标注信息；对于第一训练样本集合中的训练样本，截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像；将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到第一训练样本集合，得到新的第一训练样本集合；将新的第一训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，利用机器学习的方法，训练得到第一类别图像检测模型。
以上描述仅为本申请的较佳实施例以及对所运用技术原理的说明。本领域技术人员应当理解，本申请中所涉及的发明范围，并不限于上述技术特征的特定组合而成的技术方案，同时也应涵盖在不脱离上述发明构思的情况下，由上述技术特征或其等同特征进行任意组合而形成的其它技术方案。例如上述特征与本申请中公开的(但不限于)具有类似功能的技术特征进行互相替换而形成的技术方案。

Claims (13)

  1. 一种生成图像检测模型的方法，包括：
    获取第一训练样本集合，训练样本包括样本图像和用于表征样本图像是否为第一类别图像的标注信息；
    对于所述第一训练样本集合中的训练样本，截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像；
    将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到所述第一训练样本集合，得到新的第一训练样本集合；
    将所述新的第一训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，利用机器学习的方法，训练得到第一类别图像检测模型。
  2. 根据权利要求1所述的方法，其中，所述截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像，包括：
    将该样本图像输入预先训练的人体部位检测模型，得到检测结果信息，所述检测结果信息包括该样本图像中包括的预设人体部位的图像的位置信息，其中，所述人体部位检测模型用于表征图像和图像中包括的预设人体部位的图像的位置信息的对应关系；基于所述位置信息对该样本图像进行截取，得到新的样本图像。
  3. 根据权利要求2所述的方法，其中，所述检测结果信息还包括：该样本图像中包括的预设人体部位的图像中显示的预设人体部位的类别信息和置信度。
  4. 根据权利要求3所述的方法，其中，所述基于得到的位置信息对该样本图像进行截取，得到新的样本图像，包括：
    按照置信度由大到小的顺序，基于得到的位置信息，从该样本图像中截取预设数量的预设人体部位的图像，以及将截取的预设人体部位的图像作为新的样本图像。
  5. 根据权利要求2-4中任一所述的方法，其中，所述人体部位检测模型通过以下步骤训练得到：
    获取第二训练样本集合，训练样本包括样本图像和样本图像的标注信息，其中，标注信息包括样本图像中包含的预设人体部位的图像的位置信息和样本图像中显示的预设人体部位的类别信息；
    将所述第二训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，训练得到人体部位检测模型。
  6. 一种图像检测方法，包括：
    获取待检测图像；
    将所述待检测图像输入第一类别图像检测模型，得到用于表征所述待检测图像是否为第一类别图像的检测结果信息，其中，所述第一类别图像检测模型是按照如权利要求1-5之一所述的方法生成的。
  7. 一种生成图像检测模型的装置，包括：
    第一训练样本集合获取单元，被配置成获取第一训练样本集合，训练样本包括样本图像和用于表征样本图像是否为第一类别图像的标注信息；
    截取单元，被配置成对于所述第一训练样本集合中的训练样本，截取该训练样本的样本图像中包括的预设人体部位的图像，得到新的样本图像；
    样本添加单元，被配置成将得到的新的样本图像和新的样本图像的标注信息作为新的训练样本添加到所述第一训练样本集合，得到新的第一训练样本集合；
    训练单元，被配置成将所述新的第一训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，利用机器学习的方法，训练得到第一类别图像检测模型。
  8. 根据权利要求7所述的装置，其中，所述样本添加单元进一步被配置成：
    将该样本图像输入预先训练的人体部位检测模型，得到检测结果信息，所述检测结果信息包括该样本图像中包括的预设人体部位的图像的位置信息，其中，所述人体部位检测模型用于表征图像和图像中包括的预设人体部位的图像的位置信息的对应关系；基于所述位置信息对该样本图像进行截取，得到新的样本图像。
  9. 根据权利要求8所述的装置，其中，所述检测结果信息还包括：
    该样本图像中包括的预设人体部位的图像中显示的预设人体部位的类别信息和置信度。
  10. 根据权利要求9所述的装置，其中，所述样本添加单元进一步被配置成：
    按照置信度由大到小的顺序，基于得到的位置信息，从该样本图像中截取预设数量的预设人体部位的图像，以及将截取的预设人体部位的图像作为新的样本图像。
  11. 根据权利要求8-10中任一所述的装置，其中，所述人体部位检测模型通过以下步骤训练得到：
    获取第二训练样本集合，训练样本包括样本图像和样本图像的标注信息，其中，标注信息包括样本图像中包含的预设人体部位的图像的位置信息和样本图像中显示的预设人体部位的类别信息；
    将所述第二训练样本集合中的训练样本的样本图像作为输入，将与输入的样本图像对应的标注信息作为期望输出，训练得到人体部位检测模型。
  12. 一种电子设备，包括：
    一个或多个处理器；
    存储装置，其上存储有一个或多个程序，
    当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如权利要求1-6中任一所述的方法。
  13. 一种计算机可读介质，其上存储有计算机程序，其中，所述程序被处理器执行时实现如权利要求1-6中任一所述的方法。
PCT/CN2018/116337 2018-07-06 2018-11-20 生成图像检测模型的方法和装置 WO2020006963A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810734679.5A CN108985208A (zh) 2018-07-06 2018-07-06 生成图像检测模型的方法和装置
CN201810734679.5 2018-07-06

Publications (1)

Publication Number Publication Date
WO2020006963A1 true WO2020006963A1 (zh) 2020-01-09

Family

ID=64536251

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/116337 WO2020006963A1 (zh) 2018-07-06 2018-11-20 生成图像检测模型的方法和装置

Country Status (2)

Country Link
CN (1) CN108985208A (zh)
WO (1) WO2020006963A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111401464A (zh) * 2020-03-25 2020-07-10 北京字节跳动网络技术有限公司 分类方法、装置、电子设备及计算机可读存储介质
CN111476783A (zh) * 2020-04-13 2020-07-31 腾讯科技（深圳）有限公司 基于人工智能的图像处理方法、装置、设备及存储介质
CN111489354A (zh) * 2020-05-18 2020-08-04 国网浙江省电力有限公司检修分公司 电力塔架上鸟巢的检测方法、装置、服务器及存储介质
CN112365513A (zh) * 2020-11-23 2021-02-12 北京三快在线科技有限公司 一种模型训练的方法及装置
CN113378835A (zh) * 2021-06-28 2021-09-10 北京百度网讯科技有限公司 标注模型训练、样本标注方法及相关装置
CN113536941A (zh) * 2021-06-18 2021-10-22 浙江吉利控股集团有限公司 训练图像生成、定位检测网络训练、定位检测方法及装置
CN114324330A (zh) * 2021-12-24 2022-04-12 深圳一信泰质量技术有限公司 一种超高清智能互动显示终端性能检测装置及方法
CN117351004A (zh) * 2023-11-29 2024-01-05 杭州天眼智联科技有限公司 再生物料识别方法、装置、电子设备和计算机可读介质
CN111539443B (zh) * 2020-01-22 2024-02-09 北京小米松果电子有限公司 一种图像识别模型训练方法及装置、存储介质

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109784304B (zh) * 2019-01-29 2021-07-06 北京字节跳动网络技术有限公司 用于标注牙齿图像的方法和装置
CN109858444A (zh) * 2019-01-31 2019-06-07 北京字节跳动网络技术有限公司 人体关键点检测模型的训练方法和装置
CN110490202B (zh) * 2019-06-18 2021-05-25 腾讯科技（深圳）有限公司 检测模型训练方法、装置、计算机设备和存储介质
CN112861892B (zh) * 2019-11-27 2023-09-01 杭州海康威视数字技术股份有限公司 图片中目标的属性的确定方法和装置
CN110991312A (zh) * 2019-11-28 2020-04-10 重庆中星微人工智能芯片技术有限公司 生成检测信息的方法、装置、电子设备和介质
CN111444364B (zh) * 2020-03-04 2024-01-30 中国建设银行股份有限公司 一种图像检测方法和装置
CN111310858B (zh) * 2020-03-26 2023-06-30 北京百度网讯科技有限公司 用于生成信息的方法和装置
CN111461227B (zh) * 2020-04-01 2023-05-23 抖音视界有限公司 样本生成方法、装置、电子设备和计算机可读介质
CN112101257B (zh) * 2020-09-21 2022-05-31 北京字节跳动网络技术有限公司 训练样本生成方法、图像处理方法、装置、设备和介质
CN113269721B (zh) * 2021-04-21 2024-05-17 上海联影智能医疗科技有限公司 模型训练方法、装置、电子设备及存储介质
CN113177469B (zh) * 2021-04-27 2024-04-12 北京百度网讯科技有限公司 人体属性检测模型的训练方法、装置、电子设备及介质
CN113240376A (zh) * 2021-05-31 2021-08-10 中邮信息科技(北京)有限公司 物品信息确定方法、装置、电子设备及介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102184419A (zh) * 2011-04-13 2011-09-14 深圳市迈科龙影像技术有限公司 基于敏感部位检测的色情图像识别方法
CN103366160A (zh) * 2013-06-28 2013-10-23 西安交通大学 融合肤色、人脸和敏感部位检测的不良图像判别方法
EP2657857A1 (en) * 2012-04-27 2013-10-30 ATG Advanced Swiss Technology Group AG Method for binary classification of a query image
CN105844283A (zh) * 2015-01-16 2016-08-10 阿里巴巴集团控股有限公司 用于识别图像类目归属的方法、图像搜索方法及装置
CN107992783A (zh) * 2016-10-26 2018-05-04 上海银晨智能识别科技有限公司 人脸图像处理方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4788525B2 (ja) * 2006-08-30 2011-10-05 日本電気株式会社 物体識別パラメータ学習システム、物体識別パラメータ学習方法および物体識別パラメータ学習用プログラム
CN106548165B (zh) * 2016-11-28 2019-07-16 中通服公众信息产业股份有限公司 一种基于图像分块加权的卷积神经网络的人脸识别方法
CN107247930A (zh) * 2017-05-26 2017-10-13 西安电子科技大学 基于cnn和选择性注意机制的sar图像目标检测方法
CN107944457A (zh) * 2017-11-23 2018-04-20 浙江清华长三角研究院 一种复杂场景下的绘画物体识别与提取方法
CN108154134B (zh) * 2018-01-11 2019-07-23 天格科技（杭州）有限公司 基于深度卷积神经网络的互联网直播色情图像检测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102184419A (zh) * 2011-04-13 2011-09-14 深圳市迈科龙影像技术有限公司 基于敏感部位检测的色情图像识别方法
EP2657857A1 (en) * 2012-04-27 2013-10-30 ATG Advanced Swiss Technology Group AG Method for binary classification of a query image
CN103366160A (zh) * 2013-06-28 2013-10-23 西安交通大学 融合肤色、人脸和敏感部位检测的不良图像判别方法
CN105844283A (zh) * 2015-01-16 2016-08-10 阿里巴巴集团控股有限公司 用于识别图像类目归属的方法、图像搜索方法及装置
CN107992783A (zh) * 2016-10-26 2018-05-04 上海银晨智能识别科技有限公司 人脸图像处理方法及装置

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111539443B (zh) * 2020-01-22 2024-02-09 北京小米松果电子有限公司 一种图像识别模型训练方法及装置、存储介质
CN111401464A (zh) * 2020-03-25 2020-07-10 北京字节跳动网络技术有限公司 分类方法、装置、电子设备及计算机可读存储介质
CN111476783A (zh) * 2020-04-13 2020-07-31 腾讯科技（深圳）有限公司 基于人工智能的图像处理方法、装置、设备及存储介质
CN111489354A (zh) * 2020-05-18 2020-08-04 国网浙江省电力有限公司检修分公司 电力塔架上鸟巢的检测方法、装置、服务器及存储介质
CN111489354B (zh) * 2020-05-18 2023-07-14 国网浙江省电力有限公司检修分公司 电力塔架上鸟巢的检测方法、装置、服务器及存储介质
CN112365513A (zh) * 2020-11-23 2021-02-12 北京三快在线科技有限公司 一种模型训练的方法及装置
CN113536941A (zh) * 2021-06-18 2021-10-22 浙江吉利控股集团有限公司 训练图像生成、定位检测网络训练、定位检测方法及装置
CN113378835A (zh) * 2021-06-28 2021-09-10 北京百度网讯科技有限公司 标注模型训练、样本标注方法及相关装置
CN114324330A (zh) * 2021-12-24 2022-04-12 深圳一信泰质量技术有限公司 一种超高清智能互动显示终端性能检测装置及方法
CN114324330B (zh) * 2021-12-24 2023-09-12 深圳一信泰质量技术有限公司 一种超高清智能互动显示终端性能检测装置及方法
CN117351004A (zh) * 2023-11-29 2024-01-05 杭州天眼智联科技有限公司 再生物料识别方法、装置、电子设备和计算机可读介质
CN117351004B (zh) * 2023-11-29 2024-02-20 杭州天眼智联科技有限公司 再生物料识别方法、装置、电子设备和计算机可读介质

Also Published As

Publication number Publication date
CN108985208A (zh) 2018-12-11

Similar Documents

Publication Publication Date Title
WO2020006963A1 (zh) 生成图像检测模型的方法和装置
WO2020000879A1 (zh) 图像识别方法和装置
WO2020006964A1 (zh) 图像检测方法和装置
CN108427939B (zh) 模型生成方法和装置
US11151765B2 (en) Method and apparatus for generating information
WO2020029608A1 (zh) 用于检测电极片毛刺的方法和装置
WO2019242222A1 (zh) 用于生成信息的方法和装置
WO2020024484A1 (zh) 用于输出数据的方法和装置
CN109614934B (zh) 在线教学质量评估参数生成方法及装置
CN111414879B (zh) 人脸遮挡程度识别方法、装置、电子设备及可读存储介质
WO2020062493A1 (zh) 图像处理方法和装置
WO2020019591A1 (zh) 用于生成信息的方法和装置
WO2020056902A1 (zh) 用于处理嘴部图像的方法和装置
CN108509921B (zh) 用于生成信息的方法和装置
CN108235004B (zh) 视频播放性能测试方法、装置和系统
CN110059623B (zh) 用于生成信息的方法和装置
US11210563B2 (en) Method and apparatus for processing image
CN108229375B (zh) 用于检测人脸图像的方法和装置
CN108388889B (zh) 用于分析人脸图像的方法和装置
CN111467074B (zh) 用于检测牲畜状态的方法和装置
CN109389169A (zh) 用于处理图像的方法和装置
WO2023030427A1 (zh) 生成模型的训练方法、息肉识别方法、装置、介质及设备
JP2020013553A (ja) 端末装置に適用される情報生成方法および装置
WO2021179565A1 (zh) 用于获取信息的方法及装置
CN110059624A (zh) 用于检测活体的方法和装置

Legal Events

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

Ref document number: 18925151

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS (EPO FORM 1205A DATED 14.05.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 18925151

Country of ref document: EP

Kind code of ref document: A1