CN116342379A - Flexible and various human face image aging generation system - Google Patents

Flexible and various human face image aging generation system Download PDF

Info

Publication number
CN116342379A
CN116342379A CN202310338136.2A CN202310338136A CN116342379A CN 116342379 A CN116342379 A CN 116342379A CN 202310338136 A CN202310338136 A CN 202310338136A CN 116342379 A CN116342379 A CN 116342379A
Authority
CN
China
Prior art keywords
age
aging
condition
image
loss
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN202310338136.2A
Other languages
Chinese (zh)
Inventor
李佩佩
何召锋
王锐
曹春水
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Posts and Telecommunications
Original Assignee
Beijing University of Posts and Telecommunications
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 Beijing University of Posts and Telecommunications filed Critical Beijing University of Posts and Telecommunications
Priority to CN202310338136.2A priority Critical patent/CN116342379A/en
Publication of CN116342379A publication Critical patent/CN116342379A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/04—Context-preserving transformations, e.g. by using an importance map
    • 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
    • G06N3/0455—Auto-encoder networks; Encoder-decoder 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/18—Image warping, e.g. rearranging pixels individually
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
    • 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
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • 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
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/178—Human faces, e.g. facial parts, sketches or expressions estimating age from face image; using age information for improving recognition
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person
    • G06T2207/30201—Face
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Human Computer Interaction (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Processing (AREA)

Abstract

The invention relates to the technical field of image processing, and provides a flexible and various human face image aging generation system which comprises an acquisition unit for acquiring an original input image
Figure DDA0004157173830000011
Reference image
Figure DDA0004157173830000012
And a predefined aging textT is the same as ref The method comprises the steps of carrying out a first treatment on the surface of the A CLIP encoder for encoding the reference image
Figure DDA0004157173830000013
And the aged text t ref Mapping to the CLIP hidden space to obtain hidden vectors e respectively img Hidden vector e txt The method comprises the steps of carrying out a first treatment on the surface of the Probability age prediction unit using text prior N (e txt I) makes KL divergence constraint according to hidden vector e img Probability generation representation e of the aging condition is obtained age ＝N(μ φ (e img )，σ φ 2 (e img ) I) is carried out; diffusion self-encoder for converting an original input image
Figure DDA0004157173830000014
Encoding into semantic conditions
Figure DDA0004157173830000015
A first diffusion decoder for decoding the semantic condition z src Pre-training a noisy image that diffuses the t-th step from the encoder
Figure DDA0004157173830000016
And aging condition e age Decoding and decoding the image p subjected to denoising aging editing. By the technical scheme, the problem of low flexibility degree of face aging in the prior art is solved.

Description

Flexible and various human face image aging generation system
Technical Field
The invention relates to the technical field of image processing, in particular to a flexible and various human face image aging generation system.
Background
The aging of the face aims at keeping the identity information of the face, and simultaneously simulates the face appearance change of different age groups, and has practical landing application prospects in the aspects of age estimation, cross-age face recognition, film and television creation, medical beauty and the like. The rapid development of deep learning has driven research into human face aging over the past decades. At present, face aging mainly faces three problems: firstly, the previous GAN-based methods often have difficulty in robustly generating high quality aging results, and in the actual generation process, many results have obvious artifacts; secondly, the prior aging method usually takes a fixed age label as input, so that the flexibility of human face aging is greatly limited; finally, previous aging methods ignore the diversity of aging, as it is affected by environmental complications, and it is very unscientific to generate only one aging pattern. In summary, all three of the above problems are issues to be resolved in aging.
Disclosure of Invention
The invention provides a flexible and various human face image aging generation system, which solves the problem of low human face aging flexibility degree in the related technology.
The technical scheme of the invention is as follows: comprising the following steps:
an obtaining unit for obtaining an original input image
Figure BDA0004157173810000011
Reference image->
Figure BDA0004157173810000012
And a predefined aged text t ref ；
A CLIP encoder for encoding the reference image
Figure BDA0004157173810000013
And the aged text t ref Mapping to the CLIP hidden space to obtain hidden vectors e respectively img Hidden vector e txt ；
Probability age prediction unit using text prior N (e txt I) makes KL divergence constraint according to hidden vector e img Probability generation representation e of the aging condition is obtained age ＝N(μ φ (e img )，σ φ 2 (e img ) I) is carried out; wherein N (0,I) represents a normal distribution, mu φ Mean value and sigma of normal distribution φ Representing the variance of the normal distribution, phi being the network parameter;
diffusion self-encoder for converting an original input image
Figure BDA0004157173810000014
Encoding into semantic Condition->
Figure BDA0004157173810000015
A first diffusion decoder for decoding the semantic condition z src Pre-training a noisy image that diffuses the t-th step from the encoder
Figure BDA0004157173810000016
And aging condition e age Decoding and decoding the image p subjected to denoising aging editing.
The working principle and the beneficial effects of the invention are as follows:
because the images and the characters are used as aging conditions to be more in line with the intuition and cognition of human beings, the invention firstly refers to the images
Figure BDA0004157173810000021
And a predefined aged text t ref Mapping to the CLIP hidden space through a pre-trained CLIP encoder to obtain corresponding expression e img And e txt Utilizing the characteristic of highly consistent alignment of the CLIP hidden space text and the image; then regarding the aging condition as a sampling result from the probability distribution, using text prior as KL divergence constraint to the aging condition e age And performing probability generation representation to realize aging condition generation of flexible interaction of the image and the text.
Drawings
The invention will be described in further detail with reference to the drawings and the detailed description.
FIG. 1 is a schematic block diagram of the present invention;
FIG. 2 is a schematic diagram of a probabilistic age prediction unit according to the present invention;
FIG. 3 is a schematic diagram of a diffusion self-encoder in accordance with the present invention;
fig. 4 is a schematic diagram of an adaptive modulation module according to the present invention.
Detailed Description
The technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, the flexible and various face aging system of the present embodiment includes:
an obtaining unit for obtaining an original input image
Figure BDA0004157173810000022
Reference image->
Figure BDA0004157173810000023
And a predefined aged text t ref ；
A CLIP encoder for encoding the reference image
Figure BDA0004157173810000024
And the aged text t ref Mapping to the CLIP hidden space to obtain hidden vectors e respectively img Hidden vector e txt ；
Probability age prediction unit using text prior N (e txt I) makes KL divergence constraint according to hidden vector e img Probability generation representation e of the aging condition is obtained age ＝N(μ φ (e img )，σ φ 2 (e img ) I) is carried out; wherein N (0,I) represents a normal distribution, mu φ Mean value and sigma of normal distribution φ Representing the variance of normal distribution, phi being the network parameter of the prediction unit;
diffusion self-encoder for converting an original input image
Figure BDA0004157173810000031
Encoding into semantic Condition->
Figure BDA0004157173810000032
A first diffusion decoder for decoding the semantic condition z src Pre-training a noisy image that diffuses the t-th step from the encoder
Figure BDA0004157173810000033
And aging condition e age Decoding and decoding the image p subjected to denoising aging editing.
The present embodiment first refers to the input reference image
Figure BDA0004157173810000034
And a predefined aged text t ref Mapping to the CLIP hidden space through a pre-training CLIP encoder to respectively obtain corresponding hidden vectors e img Hidden vector e txt By utilizing the characteristic of highly consistent alignment of CLIP hidden space text and images, a lightweight network structure is designed to probability-generate an expression of aging conditions, as shown in fig. 2. The lightweight network structure first comprises a second multi-layer perceptron MLP by embedding a vector e img Inputting the average value mu obtained by the MLP of the second multi-layer perceptron φ (e img ) Sum of variances sigma φ (e img ) The aging condition is considered as the sampling result from the probability distribution: e, e age ＝N(μ φ (e img )，σ φ 2 (e img ) I) and using the text a priori N (e txt I) makes KL divergence constraints. The original input image is then ++encoded using a semantic encoder in a pre-trained diffusion self-encoder>
Figure BDA0004157173810000035
Encoding into semantic Condition z src The semantic condition includes a subject-aware feature of the image. Finally combine the semantic condition z src And age condition z obtained by aging the encoder age The first diffusion decoder is enabled to continuously execute (T steps are taken together) reverse process to obtain an image p (x) after aging editing t-1 |x t ，z tar T), where p (x T )＝N(0，I)，z tar ＝z src +z age . Wherein the first diffusion decoder employs a pre-trained diffusion decoder.
In recent years, the development of the image generation field has been greatly promoted by the proposal of DDPM (Denoising Diffusion Probabilistic Models, denoising diffusion probability model). DDPM is trained by a Forward noise adding process (Forward) and a reverse noise removing process (Denoise). In the forward noise adding process, the DDPM continuously adds noise with fixed parameters to the image to approximate to a random gaussian noise, and in the reverse process, trains the network to continuously predict the noise and makes noise reduction to restore the original image. After training, the DDPM can randomly sample from Gaussian noise to generate an image. In this embodiment, the pre-training diffusion self-Encoder trains a Semantic Encoder (Semantic Encoder) additionally on the basis of training the DDPM, and in the reverse process, the encoding characteristics of the Semantic Encoder are trained as conditions, and the structure diagram is shown in fig. 3.
It should be noted that, in the training and sampling process, the text priori may be used to make small disturbance to the text vector to directly obtain e age ＝e txt The model is beneficial to learning how to simultaneously use text and image information to guide aging generation by adopting the mode of +sigma.eta, eta-N (0,I), so that the generation of aging conditions is more flexible.
Further, an adaptive modulation module for adjusting the aging condition e age And semantic Condition z src Performing self-adaptive fusion to obtain age condition z age By age condition z age Instead of the aging condition e age Inputting the images into a decoder to obtain an aged edited image p; the step of self-adaptive fusion specifically comprises the following steps:
using a multilayer perceptron MLP to condition e aging age Mapping to diffusion self-encoder hidden space to obtain mapping vector delta z age ；
Mapping vector Deltaz by two layers of full-connection layers respectively age Learning weight parameter gamma θ And beta θ By means of adaptive coding and semantic condition z src Fusion to obtain age condition z age 。
In this embodiment, e in the CLIP hidden space is modulated adaptively age Mapping to a pre-training steganography space that diffuses from the encoder, thereby embedding e in CLIP steganography space age And semantic Condition z src Performing self-adaptive fusion to further obtain diversified aging conditions z age 。
Further, the method further comprises the following steps:
a calculation unit for performing calculation of the loss function L, specifically, the present embodiment proposes training of six loss constraint models, i.e., l=l tKL +λ 1 L age +λ 2 L clip +λ 3 L id +λ 4 L norm +λ 5 L rec 。
In this embodiment, the denoising process is performed with a pre-trained conditional diffusion decoder (Diffusion Decoder) in the diffusion self-encoder, at z src Is guided to reconstruct the original input image
Figure BDA0004157173810000041
Similarly we can reconstruct the original input image +.>
Figure BDA0004157173810000042
In order to let the obtained aging conditions meet the text prior and avoid collapsing to a fixed value, we propose a text-guided KL divergence constraint: l (L) tKL ＝D KL (N(μ φ (e img )，σ φ 2 (e img )I)||N(e txt ，I))
In the actual training process, on the premise that the hidden space of the CLIP is a hypersphere, the constraint of Euclidean distance and the constraint of negative cosine similarity are equivalent, so that the distance item in KL divergence is weakened into the constraint of negative cosine similarity and a modulus value.
To ensure that the ageing results meet the target age condition, we propose two losses: age characteristic contrast loss L age And CLIP direction loss L clip . Wherein age loss L age Is a result of intermediate reconstruction in the feature space pair of the pre-trained age estimator f ()
Figure BDA0004157173810000043
The contrast loss of cosine similarity </DEG > is specifically defined as follows:
Figure BDA0004157173810000044
since the diffusion intermediate reconstruction result is too fuzzy, a conventional L is adopted 2 Loss is liable to cause age deviationThe difference is large, so in combination with the characteristic of the diffusion model, we propose that the age comparison loss ensures the consistency of the age of the generated result, and in the experiment, we choose m to be 0.25. Meanwhile, in order to avoid age deviation introduced by a single age estimator, we use a pre-trained large model CLIP to do additional age supervision guidance:
Figure BDA0004157173810000045
Figure BDA0004157173810000051
ΔT＝E txt (t ref )-E txt (t src )
wherein E is txt (. Cndot.) and E img (. Cndot.) represents a pre-trained CLIP encoder, t in the experiment src The text we choose is expressed as "a face".
In order to improve the retention of quality-independent age-related features of the generated image, we propose L id ，L norm And L rec Specifically, a pre-trained face recognition model R (-) is adopted, so that the identity characteristics of the model are unchanged in the aging process:
Figure BDA0004157173810000052
wherein the method comprises the steps of<·>And (3) calculating the similarity of the cosine, wherein R (-) is the characteristic representation of the characteristic space of the pre-training face recognition model. To ensure the quality of the generation, we propose a regularized term loss L for the various aging conditions norm The definition is as follows:
L norm ＝||z age || 2
in order to ensure that the age-independent characteristics of the model remain unchanged, during the training process, the reference image and the input image are randomly made to be consistent, and L is used 1 Constraint is carried out:
Figure BDA0004157173810000053
our overall objective function can be summarized as follows:
L＝L tKL +λ 1 L age +λ 2 L clip +λ 3 L id +λ 4 L norm +λ 5 L rec
wherein lambda is i Is the weight of each loss.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.

Claims (3)

1. The flexible and various human face image aging generation system is characterized by comprising:
an obtaining unit for obtaining an original input image
Figure FDA0004157173800000011
Reference image->
Figure FDA0004157173800000012
And a predefined aged text t ref ；
A CLIP encoder for encoding the reference image
Figure FDA0004157173800000013
And the aged text t ref Mapping to the CLIP hidden space to obtain hidden vectors e respectively img Hidden vector e txt ；
Probability age prediction unit using text prior N (e txt I) makes KL divergence constraint according to hidden vector e img Probability generation representation e of the aging condition is obtained age ＝N(μ φ (e img )，σ φ 2 (e img ) I) is carried out; wherein N (0,I) represents a normal distribution, mu φ Representing normal componentsMean value of cloth, sigma φ Representing the variance of the normal distribution, phi being the network parameter;
diffusion self-encoder for converting an original input image
Figure FDA0004157173800000014
Encoding into semantic Condition->
Figure FDA0004157173800000015
A first diffusion decoder for decoding the semantic condition z src Pre-training a noisy image that diffuses the t-th step from the encoder
Figure FDA0004157173800000016
And aging condition e age Decoding and decoding the image p subjected to denoising aging editing.
2. The flexible multi-face image aging generation system of claim 1, further comprising:
an adaptive modulation module for adjusting the aging condition e age And semantic Condition z src Performing self-adaptive fusion to obtain age condition z age By age condition z age Instead of the aging condition e age Inputting the images into a decoder to obtain an aged edited image p; the step of self-adaptive fusion specifically comprises the following steps:
using a multilayer perceptron MLP to condition e aging age Mapping to diffusion self-encoder hidden space to obtain mapping vector delta z age ；
Mapping vector Deltaz by two layers of full-connection layers respectively age Learning weight parameters by self-adaptive coding and semantic condition z src Fusion to obtain age condition z age 。
3. The flexible multiple face image aging generation system of claim 1, further comprising
The calculating unit is configured to perform calculation of the loss function L, specifically:
L＝L tKL +λ 1 L age +λ 2 L clip +λ 3 L id +λ 4 L norm +λ 5 L rec
wherein L is tKL Represents KL divergence constraint loss, L age Represents age characteristic contrast loss, L clip Indicating the CLIP direction loss, L id Indicating identity loss, L norm Representing regular term loss, L rec Represents a consistency loss, lambda 1 、λ 2 、λ 3 、λ 4 、λ 5 Weights for each loss;
KL divergence constraint loss L tKL The method comprises the following steps:
Figure FDA0004157173800000021
age characteristic contrast loss L age The method comprises the following steps:
Figure FDA0004157173800000022
wherein,,
Figure FDA0004157173800000023
original input image ++by diffusion decoder diffusing pre-training from encoder>
Figure FDA0004157173800000024
Reconstructing to obtain->
Figure FDA0004157173800000025
Reference picture +/by diffusion decoder diffusing pre-training from encoder>
Figure FDA0004157173800000026
Reconstructing to obtain->
Figure FDA0004157173800000027
For the target image output by the first diffusion decoder, m is a parameter,<·>calculating the similarity of the cosine;
CLIP direction loss L clip The method comprises the following steps:
Figure FDA0004157173800000028
Figure FDA0004157173800000029
ΔT＝E txt (t ref )-E txt (t src )
wherein E is txt (. Cndot.) and E img (. Cndot.) represents a pre-trained CLIP encoder, t src The text is expressed for the selected text;
identity loss L id The method comprises the following steps:
Figure FDA00041571738000000210
wherein R (-) is the characteristic representation of the characteristic space of the pre-training face recognition model;
regular term loss L norm The method comprises the following steps:
L norm ＝||z age || 2
consistency loss L rec The method comprises the following steps:
Figure FDA0004157173800000031
CN202310338136.2A 2023-03-31 2023-03-31 Flexible and various human face image aging generation system Pending CN116342379A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310338136.2A CN116342379A (en) 2023-03-31 2023-03-31 Flexible and various human face image aging generation system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310338136.2A CN116342379A (en) 2023-03-31 2023-03-31 Flexible and various human face image aging generation system

Publications (1)

Publication Number Publication Date
CN116342379A true CN116342379A (en) 2023-06-27

Family

ID=86885568

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310338136.2A Pending CN116342379A (en) 2023-03-31 2023-03-31 Flexible and various human face image aging generation system

Country Status (1)

Country Link
CN (1) CN116342379A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116542292A (en) * 2023-07-04 2023-08-04 腾讯科技（深圳）有限公司 Training method, device, equipment and storage medium of image generation model
CN118297820A (en) * 2024-03-27 2024-07-05 北京智象未来科技有限公司 Training method for image generation model, image generation method, device, equipment and storage medium

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116542292A (en) * 2023-07-04 2023-08-04 腾讯科技（深圳）有限公司 Training method, device, equipment and storage medium of image generation model
CN116542292B (en) * 2023-07-04 2023-09-26 腾讯科技（深圳）有限公司 Training method, device, equipment and storage medium of image generation model
CN118297820A (en) * 2024-03-27 2024-07-05 北京智象未来科技有限公司 Training method for image generation model, image generation method, device, equipment and storage medium

Similar Documents

Publication Publication Date Title
CN109033095B (en) Target transformation method based on attention mechanism
CN116342379A (en) Flexible and various human face image aging generation system
Gai et al. New image denoising algorithm via improved deep convolutional neural network with perceptive loss
CN113658051A (en) Image defogging method and system based on cyclic generation countermeasure network
Creswell et al. Adversarial information factorization
CN113723295A (en) Face counterfeiting detection method based on image domain frequency domain double-flow network
CN113538608B (en) Controllable figure image generation method based on generation countermeasure network
Zhao et al. CREAM: CNN-REgularized ADMM framework for compressive-sensed image reconstruction
CN115457169A (en) Voice-driven human face animation generation method and system
CN116309890A (en) Model generation method, stylized image generation method and device and electronic equipment
Uddin et al. A perceptually inspired new blind image denoising method using $ L_ {1} $ and perceptual loss
CN111564205A (en) Pathological image dyeing normalization method and device
CN114820303A (en) Method, system and storage medium for reconstructing super-resolution face image from low-definition image
US11526972B2 (en) Simultaneously correcting image degradations of multiple types in an image of a face
CN113496460B (en) Neural style migration method and system based on feature adjustment
CN117291232A (en) Image generation method and device based on diffusion model
CN117291850A (en) Infrared polarized image fusion enhancement method based on learnable low-rank representation
Jeon et al. Continuous face aging generative adversarial networks
CN114283181B (en) Dynamic texture migration method and system based on sample
CN115374854A (en) Multi-modal emotion recognition method and device and computer readable storage medium
CN115034965A (en) Super-resolution underwater image enhancement method and system based on deep learning
CN118250411B (en) Light-weight personalized face vision dubbing method
CN115496989B (en) Generator, generator training method and method for avoiding image coordinate adhesion
CN117911246B (en) Multi-mode image super-resolution reconstruction method based on structured knowledge distillation
Eswar et al. Understanding Taxonomy of Generative Models

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination