CN113160035A - Human body image generation method based on posture guidance, style and shape feature constraints - Google Patents

Human body image generation method based on posture guidance, style and shape feature constraints Download PDF

Info

Publication number
CN113160035A
CN113160035A CN202110413125.7A CN202110413125A CN113160035A CN 113160035 A CN113160035 A CN 113160035A CN 202110413125 A CN202110413125 A CN 202110413125A CN 113160035 A CN113160035 A CN 113160035A
Authority
CN
China
Prior art keywords
encoder
human body
image
style
shape
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
CN202110413125.7A
Other languages
Chinese (zh)
Inventor
卢书芳
卢富男
朱翔
寿旭峰
陶相艳
高飞
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.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
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 Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202110413125.7A priority Critical patent/CN113160035A/en
Publication of CN113160035A publication Critical patent/CN113160035A/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
    • 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
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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

Landscapes

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

Abstract

The invention discloses a human body image generation method based on posture guidance, style and shape feature constraints, which comprises the following steps: (1) collecting and acquiring source human body image IsAnd a target human body image ItCalculate its pose image Ps、PtHuman body semantic segmentation image Ss、St(ii) a (2) Construct generator G and discriminator DI、DP(ii) a (3) Will IsAnd SsInput style encoder, PtInput attitude encoder, StAn input shape encoder; inputting the obtained style characteristics, posture characteristics and shape characteristics into a decoder to obtain a virtual target human body image If(ii) a (4) Handle(Is，It) And (I)s，If) As a discriminator DIInput of (P)t，It) And (P)t，If) As a discriminator DPRespectively calculating the antagonistic losses and based on IfAnd ItCalculating image reconstruction loss, perception loss and semantic loss, and optimizing G; (5) and (5) performing iterative training to obtain a generator G for generating the human body image. By using the method and the device, style characteristics can be extracted according to the semantic region, and the posture and the shape of the human body image can be controlled.

Description

Human body image generation method based on posture guidance, style and shape feature constraints
Technical Field
The invention belongs to the technical field of human body image generation, and particularly relates to a human body image generation method based on posture guidance, style and shape feature constraints.
Background
Human body image generation is an important branch in the field of computer vision, and can be widely applied to the fields of data enhancement of pedestrian re-identification, movie role making, virtual fitting, augmented reality and the like. The human body image generation based on the posture guidance means that a target posture and a (group of) source image are given, and under the guidance of the target posture, a target human body image with the style characteristics of the source image in the target posture is generated.
For example, chinese patent publication No. CN112116673A discloses a method for generating a virtual human body image based on structural similarity under posture guidance; chinese patent publication No. CN109191366A discloses a method and apparatus for synthesizing multi-view human body images based on human body posture.
The current human body image generation has two problems: (1) in the style feature extraction, a global style feature is often extracted by taking the whole source image as an input, and the feature of a specific semantic area cannot be extracted independently. (2) The control mode is single, only the posture of the source image can be changed, and the style and the shape of a specific semantic area cannot be controlled.
Therefore, there is a need for a human body image generation method that can provide a variety of image synthesis control methods.
Disclosure of Invention
The invention provides a human body image generation method based on posture guidance, style and shape feature constraints, which can extract style features according to semantic regions and control the posture and the shape of a human body image.
A human body image generation method based on posture guidance, style and shape feature constraints comprises the following steps:
(1) collecting and acquiring source human body image IsAnd a target human body image ItRespectively obtaining a posture image P of the source human body image and the target human body image according to the two imagess、PtHuman body semantic segmentation image Ss、St；
(2) Construct generator G and discriminator DI、DPWherein the generator G comprises a stylized Encoder EncoderstyleEncoder for gesture coderposeEncoder of shape EncodershapeAnd a Decoder; discriminator DIFor discriminating virtual target image IfAnd source human body image IsThe texture similarity between them; discriminator DPFor discriminating virtual target image IfWith the target attitude PtThe consistency of (2);
(3) the source human body image I obtained in the step (1) is processedsSource-fused human body semantic segmentation image SsInput style Encoder EncoderstyleObject pose image PtInput attitude Encoder EncoderposeTarget human semantic segmentation image StInput shape Encoder Encodershape；
Inputting the style characteristic, the posture characteristic and the shape characteristic which are extracted in sequence into a Decoder to obtain a virtual target human body image If；
(4) Handle (I)s，It) And (I)s，If) Respectively as a discriminator DIInput of (P)t，It) And (P)t，If) Respectively as a discriminator DPRespectively calculating the opposing loss LadvAnd is based on IfAnd ItCalculating image reconstruction loss LreconstructionAnd a loss of perception LperceptualAnd semantic loss LCXG is optimized;
(5) and (5) circulating the step (3) and the step (4), and obtaining a trained generator G after the preset iteration times are reached, and using the generator G for generating the virtual target image in the real scene.
In the step (1), the number N of key points of the pose image is 18, and the number C of classes of the human semantic segmentation image is 8.
The specific steps of the step (2) are as follows:
(2-1) construction of stylistic Encoder Encoderstyle
EncoderstyleThe VGG network comprises 5 pre-trained VGG networks of 3 x 3 convolutional layers, and the sizes of feature maps extracted from the first 4 convolutional layers respectively correspond to the sizes of {1_1,2_1,3_1,4_1} feature maps in the VGG; combining the features extracted by the convolutional layer and the features extracted by the VGG network in sequence, and inputting the next convolutional layer; a last convolutional layer, mapping features from 1024 dimensions to 64 dimensions;
when in use, firstly, the semantic segmentation image is used for segmenting 8 independent images
Figure BDA0003024725740000021
Then 8 independent semantic images are respectively input into the EncoderstyleAnd finally, sequentially cascading the style features to obtain the final 512-dimensional style feature. (2-2) construction of an attitude Encoder EncoderposeAnd shape Encoder Encodershape
EncoderposeAnd EncodershapeThe network structures of the devices are the same, and the devices all comprise 4 3 multiplied by 3 convolution layers, wherein the activation layer is a ReLU layer, and 512-dimensional posture features and shape features are extracted;
(2-3) construction of Decoder
Taking the attitude characteristics as input, and calculating a normalization parameter by using the style characteristics and the shape characteristics; firstly, 4 ResBlock are passed through, and the channel is kept unchanged; then 3 groups of upper sampling layers and ResBlock layers are passed; the remaining active layers are the ReLU layers, except for the last active layer, which is tan h.
(2-4) construction of the discriminator DI、DP
Using PatchGAN as the discriminator, including 4 3 × 3 convolutional layers and 3 residual blocks, Dropout of the discriminator is set to 0.5.
In step (4), the countermeasure loss function is defined as:
Figure BDA0003024725740000031
in the formula, E represents a desirable value.
In step (4), the image reconstruction loss LreconstructionIs L between the virtual target image and the real target image1The loss, defined as:
Lreconstruction＝||G(Is，Ss，Pt，St)-It||1.
the image perception loss is defined as:
Figure BDA0003024725740000032
Figure BDA0003024725740000033
wherein the content of the first and second substances,
Figure BDA0003024725740000034
a gram matrix is represented that is,
Figure BDA0003024725740000035
representing I extracted with a pre-trained VGG19 networktLayer i profile, i ═ relu {3_2,4_2 };
loss of semantics LCXIs defined as:
Figure BDA0003024725740000036
in the formula (I), the compound is shown in the specification,
Figure BDA0003024725740000041
representing I extracted with a pre-trained VGG19 networkfThe characteristic map of the l-th layer. In the step (5), in the training process, the learning rate is initially 0.0001, and in 1000 iterations, the linear attenuation is to 0.
Compared with the prior art, the invention has the following beneficial effects:
1. in the human body image generation method based on the posture guidance, the style and the shape feature constraint, the style encoder based on the semantic segmentation image can independently extract the features of each semantic region and combine the features into style features according to the preset sequence, so that the features between different semantic regions have independence, the style feature recombination can be realized under the condition of a group of source images, and the method is more flexible in practical application.
2. In the human body image generation method based on the posture guidance, the style and the shape feature constraint, the decoder uses the shape feature of the target semantic segmentation image for normalization, and can output the image which is consistent with the target semantic segmentation.
Drawings
FIG. 1 is a schematic flow diagram of the process of the present invention;
FIG. 2 is a schematic diagram of human body image pose according to the present invention;
FIG. 3 is a schematic diagram of human body image semantic segmentation according to the present invention;
FIG. 4 is a schematic diagram of a stylized encoder of the present method.
Detailed Description
The invention will be described in further detail below with reference to the drawings and examples, which are intended to facilitate the understanding of the invention without limiting it in any way.
As shown in fig. 1, a human body image generation method based on posture guidance, style and shape feature constraints includes the following steps:
step 1, acquiring and acquiring a source human body image IsAnd a target human body image It(ii) a Respectively obtaining a posture image P of the source human body image and a target human body image according to the source human body image and the target human body images，PtHuman body semantic segmentation image Ss，St。
Specifically, as shown in fig. 2, the number N of pose image key points is 18; as shown in fig. 3, the number of classes C of human semantic segmentation images is 8.
Step 2, constructing a generator G and a discriminator DI、DPWherein the generator G comprises a stylized Encoder EncoderstyteEncoder for gesture coderposeEncoder of shape EncodershapeAnd a Decoder.
The method comprises the following specific steps:
step 2.1, constructing EncoderstyleAs shown in fig. 4.
EncoderstvleThe VGG network comprises 5 pre-trained VGG networks of 3 x 3 convolutional layers, and the sizes of feature maps extracted from the first 4 convolutional layers respectively correspond to the sizes of {1_1,2_1,3_1,4_1} feature maps in the VGG. And combining the features extracted by the convolutional layer and the features extracted by the VGG network in sequence, and inputting the next convolutional layer. The last convolutional layer maps features from 1024 dimensions to 64 dimensions.
When in use, firstly, the semantic segmentation image is used for segmenting 8 independent images
Figure BDA0003024725740000051
Then 8 independent semantic images are respectively input into the EncoderstvleAnd finally, sequentially cascading the style features to obtain the final 512-dimensional style feature.
Step 2.2, constructing EncoderposeAnd Encodershape
EncoderposeAnd EncodershapeThe network structures of the three-dimensional space network are the same, the three-dimensional space network comprises 4 3 x 3 convolution layers, the active layer is a ReLU layer, and 512-dimensional posture features and shape features are extracted.
Step 2.3, constructing the Decoder
With the pose features as input, normalization parameters are calculated using the style features and the shape features.
Firstly, 4 ResBlock are passed through, and the channel is kept unchanged; then go through 3 sets of upsampled and ResBlock layers, all except the last active layer being tanh, the remaining active layers being ReLU layers.
Step 3.4, constructing a discriminator DI、DP
Using PatchGAN as the discriminator, including 4 3 × 3 convolutional layers and 3 residual blocks, Dropout of the discriminator is set to 0.5.
Step 3, the source human body image I obtained in the step 1 is processedsSource-fused human body semantic segmentation image SsInput style Encoder EncoderstyleObject pose image PtInput attitude Encoder EncoderposeTarget human semantic segmentation image StInput shape Encoder Encodershape(ii) a Inputting the style characteristic, the posture characteristic and the shape characteristic which are extracted in sequence into a Decoder to obtain a virtual target human body image If。
Step 4, treating (I)s，It) And (I)s，If) Respectively as a discriminator DIInput of (P)t，It) And (P)t，If) Respectively as a discriminator DPRespectively calculating the opposing loss LadvAnd is based on IfAnd ItCalculating image reconstruction loss LreconstructionAnd a loss of perception LperceptualAnd semantic loss LCXAnd G is optimized.
Specifically, the penalty function is defined as:
Figure BDA0003024725740000061
wherein E represents expectation.
Image reconstruction loss is L between the virtual target image and the real target image1The loss, defined as:
Lreconstruction＝||G(Is，Ss，Pt，St)-It||1.
the image perception loss is defined as:
Figure BDA0003024725740000062
Figure BDA0003024725740000063
wherein
Figure BDA0003024725740000064
A gram matrix is represented that is,
Figure BDA0003024725740000065
representing I extracted with a pre-trained VGG19 networktI ═ relu {3_2,4_2 }.
Loss of semantics LCXIs defined as:
Figure BDA0003024725740000066
and 5, circulating the step 3 and the step 4, and obtaining a trained generator G after the preset iteration times are reached, wherein the generator G is used for generating the virtual target image in the real scene.
Specifically, in the training process, the learning rate is initially 0.0001, and in 1000 iterations, the linear decay is to 0.
The embodiments described above are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only specific embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions and equivalents made within the scope of the principles of the present invention should be included in the scope of the present invention.

Claims (6)

1. A human body image generation method based on posture guidance, style and shape feature constraints is characterized by comprising the following steps:
(1) collecting and acquiring source human body image IsAnd a target human body image ItRespectively obtaining a posture image P of the source human body image and the target human body image according to the two imagess、PtHuman body semantic segmentation image Ss、St；
(2) Construct generator G and discriminator DI、DPWherein the generator G comprises a stylized Encoder EncoderstyleEncoder for gesture coderposeEncoder of shape EncodershapeAnd a Decoder; discriminator DIFor discriminating virtual target image IfAnd source human body image IsThe texture similarity between them; discriminator DPFor discriminating virtual target image IfWith the target attitude PtThe consistency of (2);
(3) the source human body image I obtained in the step (1) is processedsSource-fused human body semantic segmentation image SsInput style Encoder EncoderstyleObject pose image PtInput attitude Encoder EncoderposeTarget human semantic segmentation image StInput shape Encoder Encodershape；
Inputting the style characteristic, the posture characteristic and the shape characteristic which are extracted in sequence into a Decoder to obtain a virtual target human body image If；
(4) Handle (I)s,It) And (I)s,If) Respectively as a discriminator DIInput of (P)t，It) And (P)t，If) Respectively as a discriminator DPRespectively calculating the opposing loss LadvAnd is based on IfAnd ItCalculating image reconstruction loss LreconstructionAnd a loss of perception LperceptualAnd semantic loss LCXG is optimized;
(5) and (5) circulating the step (3) and the step (4), and obtaining a trained generator G after the preset iteration times are reached, and using the generator G for generating the virtual target image in the real scene.
2. The method for generating a human body image based on pose guidance, style and shape feature constraints according to claim 1, wherein in step (1), the number of key points N of the pose image is 18, and the number of classes C of the human body semantic segmentation image is 8.
3. The human body image generation method based on the posture guidance, style and shape feature constraints as claimed in claim 1, wherein the specific steps of step (2) are:
(2-1) construction of stylistic Encoder Encoderstyle
EncoderstyleThe VGG network comprises 5 3 × 3 convolutional layers and a pre-trained VGG network, wherein the sizes of feature maps extracted from the first 4 convolutional layers respectively correspond to the sizes of feature maps of {1_1,2_1,3_1,4_1} layers in the VGG; combining the features extracted by the convolutional layer and the features extracted by the VGG network in sequence, and inputting the next convolutional layer; a last convolutional layer, mapping features from 1024 dimensions to 64 dimensions;
when in use, firstly, the semantic segmentation image is used for segmenting 8 independent images
Figure FDA0003024725730000022
Then 8 independent semantic images are respectively input into the EncoderstyleOutputting corresponding style characteristics, and finally cascading the style characteristics in sequence to obtain a final 512-dimensional style characteristic;
(2-2) construction of an attitude Encoder EncoderposeAnd shape Encoder Encodershape
EncoderposeAnd EncodershapeThe network structures of the devices are the same, and the devices all comprise 4 3 multiplied by 3 convolution layers, wherein the activation layer is a ReLU layer, and 512-dimensional posture features and shape features are extracted;
(2-3) construction of Decoder
Taking the attitude characteristics as input, and calculating a normalization parameter by using the style characteristics and the shape characteristics; firstly, 4 ResBlock are passed through, and the channel is kept unchanged; then 3 groups of upper sampling layers and ResBlock layers are passed; the remaining active layers are the ReLU layers, except for the last active layer, which is tan h.
(2-4) construction of the discriminator DI、DP
Using PatchGAN as the discriminator, including 4 3 × 3 convolutional layers and 3 residual blocks, Dropout of the discriminator is set to 0.5.
4. The method for generating human body image based on pose guidance, style and shape feature constraints according to claim 1, wherein in the step (4), the confrontation loss function is defined as:
Figure FDA0003024725730000021
in the formula, E represents a desirable value.
5. The human image generation method based on pose guidance, style and shape feature constraints of claim 1, wherein in step (4), image reconstruction loses LreconstructionIs L between the virtual target image and the real target image1The loss, defined as:
Lreconstruction＝||G(Is,Ss,Pt,St)-It||1.
the image perception loss is defined as:
Figure FDA0003024725730000031
wherein the content of the first and second substances,
Figure FDA0003024725730000032
a gram matrix is represented that is,
Figure FDA0003024725730000033
representing I extracted with a pre-trained VGG19 networktLayer i profile, i ═ relu {3_2,4_2 };
loss of semantics LCXIs defined as:
Figure FDA0003024725730000034
in the formula (I), the compound is shown in the specification,
Figure FDA0003024725730000035
representing I extracted with a pre-trained VGG19 networkfThe characteristic map of the l-th layer.
6. The method for generating human body image based on posture guidance, style and shape feature constraints as claimed in claim 1, wherein in step (5), during training, the learning rate is initially 0.0001, and in 1000 iterations, the linear decay is 0.
CN202110413125.7A 2021-04-16 2021-04-16 Human body image generation method based on posture guidance, style and shape feature constraints Pending CN113160035A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110413125.7A CN113160035A (en) 2021-04-16 2021-04-16 Human body image generation method based on posture guidance, style and shape feature constraints

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110413125.7A CN113160035A (en) 2021-04-16 2021-04-16 Human body image generation method based on posture guidance, style and shape feature constraints

Publications (1)

Publication Number Publication Date
CN113160035A true CN113160035A (en) 2021-07-23

Family

ID=76868198

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110413125.7A Pending CN113160035A (en) 2021-04-16 2021-04-16 Human body image generation method based on posture guidance, style and shape feature constraints

Country Status (1)

Country Link
CN (1) CN113160035A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113919998A (en) * 2021-10-14 2022-01-11 天翼数字生活科技有限公司 Image anonymization method based on semantic and attitude map guidance
CN114821811A (en) * 2022-06-21 2022-07-29 平安科技（深圳）有限公司 Method and device for generating person composite image, computer device and storage medium
CN115063536A (en) * 2022-06-30 2022-09-16 中国电信股份有限公司 Image generation method and device, electronic equipment and computer readable storage medium
CN115147508A (en) * 2022-06-30 2022-10-04 北京百度网讯科技有限公司 Method and device for training clothing generation model and method and device for generating clothing image

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113919998A (en) * 2021-10-14 2022-01-11 天翼数字生活科技有限公司 Image anonymization method based on semantic and attitude map guidance
CN113919998B (en) * 2021-10-14 2024-05-14 天翼数字生活科技有限公司 Picture anonymizing method based on semantic and gesture graph guidance
CN114821811A (en) * 2022-06-21 2022-07-29 平安科技（深圳）有限公司 Method and device for generating person composite image, computer device and storage medium
CN114821811B (en) * 2022-06-21 2022-09-30 平安科技（深圳）有限公司 Method and device for generating person composite image, computer device and storage medium
CN115063536A (en) * 2022-06-30 2022-09-16 中国电信股份有限公司 Image generation method and device, electronic equipment and computer readable storage medium
CN115147508A (en) * 2022-06-30 2022-10-04 北京百度网讯科技有限公司 Method and device for training clothing generation model and method and device for generating clothing image
CN115147508B (en) * 2022-06-30 2023-09-22 北京百度网讯科技有限公司 Training of clothing generation model and method and device for generating clothing image
CN115063536B (en) * 2022-06-30 2023-10-10 中国电信股份有限公司 Image generation method, device, electronic equipment and computer readable storage medium

Similar Documents

Publication Publication Date Title
CN113160035A (en) Human body image generation method based on posture guidance, style and shape feature constraints
CN109255831B (en) Single-view face three-dimensional reconstruction and texture generation method based on multi-task learning
CN111340814B (en) RGB-D image semantic segmentation method based on multi-mode self-adaptive convolution
CN113077471A (en) Medical image segmentation method based on U-shaped network
CN115482241A (en) Cross-modal double-branch complementary fusion image segmentation method and device
CN113496507A (en) Human body three-dimensional model reconstruction method
CN114627006B (en) Progressive image restoration method based on depth decoupling network
CN113255813B (en) Multi-style image generation method based on feature fusion
CN113362422B (en) Shadow robust makeup transfer system and method based on decoupling representation
CN111986075A (en) Style migration method for target edge clarification
CN112686898A (en) Automatic radiotherapy target area segmentation method based on self-supervision learning
CN112686816A (en) Image completion method based on content attention mechanism and mask code prior
CN111462274A (en) Human body image synthesis method and system based on SMP L model
CN112084934A (en) Behavior identification method based on two-channel depth separable convolution of skeletal data
CN113658040A (en) Face super-resolution method based on prior information and attention fusion mechanism
CN113538608A (en) Controllable character image generation method based on generation countermeasure network
KR102419270B1 (en) Apparatus and method for segmenting medical image using mlp based architecture
CN115619933A (en) Three-dimensional face reconstruction method and system based on occlusion segmentation
DE102021124537A1 (en) ENERGY-BASED VARIATIONAL AUTOENCODER
CN113222808A (en) Face mask removing method based on generative confrontation network
CN117557683A (en) Novel image generation method driven by text and semantic segmentation map together
CN114972619A (en) Single-image face three-dimensional reconstruction method based on self-alignment double regression
CN116777738A (en) Authenticity virtual fitting method based on clothing region alignment and style retention modulation
CN116152060A (en) Double-feature fusion guided depth image super-resolution reconstruction method
CN115482368A (en) Method for editing three-dimensional scene by utilizing semantic graph

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