CN110942025A - Unsupervised cross-domain pedestrian re-identification method based on clustering - Google Patents
Unsupervised cross-domain pedestrian re-identification method based on clustering Download PDFInfo
- Publication number
- CN110942025A CN110942025A CN201911172875.9A CN201911172875A CN110942025A CN 110942025 A CN110942025 A CN 110942025A CN 201911172875 A CN201911172875 A CN 201911172875A CN 110942025 A CN110942025 A CN 110942025A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- picture
- domain
- label
- model
- 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.)
- Withdrawn
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
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (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)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an unsupervised cross-domain pedestrian re-identification method based on clustering, which comprises the steps of firstly, inputting a source domain picture with a label into a defined network architecture for pre-training to obtain a baseline model; secondly, inputting the style-converted label-free target domain image into a baseline model for feature extraction and defining a pseudo label, and refining the pre-trained model of the previous stage by using the defined pseudo label; and finally, loading the trained pedestrian re-recognition model, extracting the pedestrian picture characteristics of the picture to be retrieved and the target domain, retrieving the best matched pedestrian picture from the target domain and outputting the pedestrian picture. The invention effectively improves the practicability of the pedestrian re-identification model in actual life, improves the re-identification performance, and has good network performance and strong generalization capability.
Description
Technical Field
The invention belongs to the technical field of image recognition, and particularly relates to an unsupervised cross-domain pedestrian re-recognition method based on clustering.
Background
In recent years, pedestrian re-identification has been widely studied in the field of computer vision, and the aim is to search and output a person in videos shot by several cameras which do not overlap with each other, given a picture of the pedestrian to be searched. The existing pedestrian re-identification method is based on a prior condition: the pedestrians in all the pictures are detected by the detection frame, and the target domain data set is the image of the pedestrian framed by the detection frame. The initial approach relied on manual extraction of features to label the data set, which was not only time consuming and laborious but also consistently low performance. With the rapid development of deep learning in recent years, the performance of pedestrian re-identification is greatly improved. The deep learning has the advantages that the convolutional neural network can automatically extract pedestrian features, calculate the distance between the features, judge the similarity and enhance the robustness of the model through repeated iterative training. The initial pedestrian re-identification method based on deep learning mainly focuses on global features, but the global features usually ignore some local information important in identification, such as shoes, clothes logo, sunglasses and the like. Therefore, researchers have proposed a method based on local features, which is mainly divided into two aspects: one is a component-based approach, focusing on locating regions with specific semantics for local representation; the other method is a method of directly dividing the picture horizontally. Recently, a method for generating and expanding a data set based on a GAN network is widely researched, mainly used for solving the problem of style difference among different cameras, and also applied to expanding the data set based on human posture change, but human posture estimation is mostly detected according to key points, and due to the fact that pictures shot by the cameras have the problems of alignment, shielding, light and the like, the posture of a human body cannot be accurately estimated.
Most of the above mentioned learning strategies based on supervision are supervised training on a certain data set, and the training effect is better on the data set, but the performance is greatly degraded when the data set is migrated to other data sets. In addition, in actual life, labeling of a large number of data sets is time-consuming and labor-consuming, and basically cannot be realized, so that recently, researchers propose to use an unsupervised domain self-adaptive method to improve the performance of a pedestrian re-identification model on a label-free target training set. Unsupervised here means that the target data set is unlabeled, but the reference data set may be labeled.
Disclosure of Invention
The purpose of the invention is as follows: the invention provides an unsupervised cross-domain pedestrian re-identification method based on clustering, which effectively improves the practicability of a pedestrian re-identification model in actual life, improves the re-identification performance, and has good network performance and strong generalization capability.
The technical scheme is as follows: the invention relates to an unsupervised cross-domain pedestrian re-identification method based on clustering, which comprises the following steps of:
(1) inputting the source domain picture with the label into a self-defined network model for pre-training to obtain a baseline model;
(2) inputting the target domain picture without the label into a StarGAN network to generate a new picture with the invariance of a camera, and forming a new training set with the target domain picture;
(3) preprocessing the new training set, and inputting the new training set into the baseline model trained in the step (1) to extract features serving as final feature representation of each image;
(4) calculating distances pairwise according to the eigenvectors generated in the step (3), and defining pseudo labels for a part of pedestrian images according to a self-defined threshold;
(5) clustering the target domain images according to the label defined in the step (4) in combination with a clustering algorithm, and re-labeling the target domain images according to a clustering result;
(6) using a self-defined pseudo label as a baseline model obtained by training in the step (1) of triple loss refinement of the supervision information and the difficultly-divided sample, and re-labeling the pseudo label on the pedestrian every iteration until the pedestrian is stable;
(7) and loading the trained pedestrian re-recognition model, inputting the picture to be retrieved, extracting the picture to be retrieved and the pedestrian picture characteristics of the target domain, retrieving the best matched pedestrian picture and outputting the pedestrian picture.
Further, the customized network model in step (1) is as follows:
using ResNet50 pre-trained on ImageNet as a main network, reserving the part before res _ conv _4_2 of the main network, dividing the network after the main network into two branches, wherein the two branches have similar structures but different down-sampling rates and comprise a global branch and a local branch, the global branch learns the overall feature representation, the local branch emphasizes the feature extraction of details, and the triple loss and the softmax loss are used for refining the baseline model.
Further, the preprocessing in the step (3) mainly includes using random erasing, random flipping and random clipping techniques.
Further, the final feature representation of the image in the step (3) is obtained by stitching the global feature vector and the local feature vector.
Further, the step (4) is realized as follows:
calculating the distance between the characteristic vectors by adopting Euclidean distance, and defining a label as y for a certain image0And according to the distance between the residual unmarked pictures and the images of the defined pseudo labels and a self-defined threshold value, if the distance between a certain image and the defined pseudo labels is greater than the threshold value, marking the image as the pseudo label, otherwise, defaulting that the image and a certain picture in the defined pseudo labels belong to the same pedestrian.
Further, the loss function of step (6) is as follows:
and randomly extracting P identities and K instances of each small batch to meet the requirement of loss of the batch-hard triple:
the method comprises the following steps that a hardest positive sample in a triple is selected as a sample which is farthest away from an anchor in the same class, and a hardest negative sample in a non-same class is selected as a sample which is closest to the anchor;
the loss function for camera invariance is:
wherein n istIs the number of real target pictures in the training batch,
is the number of samples generated by the camera style migration,
c is the number of cameras; the overall loss function is:
where β ∈ (0,1) is the weight of the loss of camera invariance.
Has the advantages that: compared with the prior art, the invention has the beneficial effects that: 1. the global and local pedestrian characteristic information is considered in the baseline model designed in the clustering-based pedestrian re-identification method, and the designed network has higher robustness; 2. according to the method, the style difference of the images shot by different cameras is considered, the style of the target domain image is converted, and the performance of the model is improved while data is enhanced; 3. the invention adopts a clustering method to endow the label-free target domain image with a pseudo label as the supervision information to refine the model pre-trained on the source data set, thereby improving the cross-domain performance of the model and having strong generalization capability.
Drawings
FIG. 1 is a block flow diagram of a pedestrian re-identification method according to the present invention;
FIG. 2 is a basic architecture of the baseline model of the present invention;
FIG. 3 is a model refinement process based on clustering on a target domain according to the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings, as shown in fig. 1, the present invention specifically includes the following steps:
the method comprises the following steps: and inputting the source domain picture with the label into a designed network model for pre-training to obtain a baseline model.
As shown in fig. 2: using ResNet50 pre-trained on ImageNet as a backbone network, reserving the part before res _ conv _4_2 of the backbone network, dividing the latter part into two independent branches, sharing an architecture similar to that of the original ResNet-50, wherein the two branches have similar structures but different down-sampling rates and comprise a global branch and a local branch, the global branch emphasizes the overall feature representation, the global branch removes the last fully-connected layer of the original ResNet-50, two new fully-connected layers are added, the dimension of the first layer is 2048 dimensions, called FC- #2048, the dimension of the second layer is the number of pedestrian IDs in the source data set, called FC- # ID, pedestrian re-identification is converted to classification tasks using softmax loss after FC- # ID, meanwhile, applying triplet loss on FC- #2048 converts pedestrian re-identification into an authentication task, wherein the softmax loss function is as follows:
wherein N is the size of the mini-batch in the training process, and C is the number of the pedestrian IDs in the source training set.
The triple loss function is defined as:
in the above formula, P pedestrian IDs of training samples and K samples corresponding to each ID are randomly selected,
the features of anchor, positive and negative respectively represent pedestrian IDs, and i, j in the superscript represent pedestrian IDs respectively.
The local branch focuses on more detailed feature extraction, the same convolution operation as the global branch is carried out on the feature map obtained by res _ conv _4_1, a down-sampling operation is not adopted during convolution, then the feature map obtained after convolution is horizontally divided into an upper part and a lower part, global average pooling operation is respectively carried out, dimension reduction operation is carried out by utilizing a full connection layer, two 256-dimensional feature vectors are obtained by utilizing the full connection layer and are respectively used as feature representations of the upper half body and the lower half body of a pedestrian, and a baseline model with good performance is obtained by applying softmax loss to carry out optimization training on the model.
By combining global and local information, a more robust and discriminative representation of pedestrian features can be learned.
As shown in FIG. 3, the model refinement process based on clustering on the target domain of the present invention is as follows:
step two: and inputting the target domain picture without the label into the StarGAN network to generate the target domain picture with the invariableness of the camera.
In the re-recognition task, pictures come from different cameras and the camera IDs are known, the performance of a pedestrian re-recognition model is greatly reduced due to the difference of shooting styles of the different cameras, in order to solve the problem, each camera is regarded as a style domain, the style-converted pictures are generated by using StarGAN, the style-converted pictures and an original training sample form an enhanced training set, and due to interference factors such as shielding, illumination, partial deletion and the like in the images, in order to learn more robust feature representation, data augmentation is carried out by adopting random erasing, random overturning and random clipping technologies.
Step three: and inputting the target domain picture and the picture generated in the second step into the baseline model trained in the first step to extract features as final representation of each image.
Since StarGAN cannot model the transmission process perfectly, errors occur in the image generation process, and due to occlusion and detection errors, noise samples exist in real data, and converting the noise samples into false data may generate more noise samples, so the source data set and the generated new picture are adopted as a training set in a ratio of 3: 1. The trained baseline model in the step one can achieve a feature extraction function with better performance, so that the training set is input into the pre-trained baseline model in the step one for pedestrian feature extraction, and in the step, the global feature vector and the local feature vector are spliced to be used as final feature representation of the pedestrian, so that the distance between pedestrians can be conveniently calculated, and a pseudo label is defined for the pedestrian.
Step four: and e, calculating the distance pairwise according to the feature vectors generated in the step three, and defining a pseudo label for a part of images according to a self-defined threshold value.
Calculating the distance between each feature vector by Euclidean distance, and firstly defining a label as y for a certain image0According to the distance between the residual unmarked picture and the image with defined pseudo labelAnd (4) a distance and a self-defined threshold value, and if the distance between a certain image and the defined pseudo label is larger than the threshold value, marking the image as the pseudo label (from y)0Start increasing in order), otherwise the image and a certain photo in the defined pseudo label belong to the same pedestrian by default.
The normalized euclidean distance formula is defined as follows:
first, 0-mean normalization is performed on each vector, and the formula is as follows:
where σ is the sample standard deviation.
The euclidean distance between the normalized feature vectors is then calculated:
vector f (f)1,f2,...,fn)g:(g1,g2,...,gn)。
Step five: and defining a pseudo label for the target domain image according to the label defined in the last step and a clustering algorithm.
And using the number of the images with the defined labels in the previous step as a K value of a K-means algorithm, clustering the target domain image set by using the idea of the K-means algorithm, and re-labeling the target domain images according to the clustering result. The pseudo-code for the K-means algorithm is represented as follows:
in the invention, the value of K is the number of pictures which are endowed with pseudo labels in the previous step, and the data of all target domains are endowed with pseudo labels again through a K-means algorithm.
Step six: and (3) refining the trained model until the trained model is stable by using the self-defined pseudo label as the supervision information and the nondifferential sample triple loss, and noting that the pseudo label is re-labeled to the pedestrian every iteration.
And randomly extracting P identities and K instances of each small batch to meet the requirement of loss of the batch-hard triple:
the candidate of hardest positive is selected as the sample with the farthest distance from the anchor in the same class, and the candidate of hardest negative is selected as the sample with the nearest distance from the anchor in the non-same class. Specifically, given a picture, the feature vector and the custom pseudo tag of each picture are used as the two inputs to the triplet.
Considering the loss of camera invariance of StarGAN, we consider that each image and the corresponding image after style conversion have the same identity, so the loss function of camera invariance in this invention can be expressed as:
in the formula, ntIs the number of real target pictures in the training batch,
is the number of samples generated by the camera style migration,
c is the number of cameras. In this way, images of the same ID are closer together under different cameras.
Therefore, the total loss function of step six in the present invention can be expressed as:
where β ∈ (0,1) is the weight of the loss of camera invariance.
Step seven: and loading the trained pedestrian re-recognition model, inputting the picture to be retrieved, extracting the picture to be retrieved and the pedestrian picture characteristics of the target domain, retrieving the best matched pedestrian picture and outputting the pedestrian picture.
Claims (6)
1. An unsupervised cross-domain pedestrian re-identification method based on clustering is characterized by comprising the following steps:
(1) inputting the source domain picture with the label into a self-defined network model for pre-training to obtain a baseline model;
(2) inputting the target domain picture without the label into a StarGAN network to generate a new picture with the invariance of a camera, and forming a new training set with the target domain picture;
(3) preprocessing the new training set, and inputting the new training set into the baseline model trained in the step (1) to extract features serving as final feature representation of each image;
(4) calculating distances pairwise according to the eigenvectors generated in the step (3), and defining pseudo labels for a part of pedestrian images according to a self-defined threshold;
(5) clustering the target domain images according to the label defined in the step (4) in combination with a clustering algorithm, and re-labeling the target domain images according to a clustering result;
(6) using a self-defined pseudo label as a baseline model obtained by training in the step (1) of triple loss refinement of the supervision information and the difficultly-divided sample, and re-labeling the pseudo label on the pedestrian every iteration until the pedestrian is stable;
(7) and loading the trained pedestrian re-recognition model, inputting the picture to be retrieved, extracting the picture to be retrieved and the pedestrian picture characteristics of the target domain, retrieving the best matched pedestrian picture and outputting the pedestrian picture.
2. The unsupervised cluster-based cross-domain pedestrian re-identification method according to claim 1, wherein the customized network model in step (1) is as follows:
using ResNet50 pre-trained on ImageNet as a main network, reserving the part before res _ conv _4_2 of the main network, dividing the network after the main network into two branches, wherein the two branches have similar structures but different down-sampling rates and comprise a global branch and a local branch, the global branch learns the overall feature representation, the local branch emphasizes the feature extraction of details, and the triple loss and the softmax loss are used for refining the baseline model.
3. The unsupervised cluster-based cross-domain pedestrian re-identification method according to claim 1, wherein the preprocessing of step (3) mainly comprises using random erasure, random flipping and random clipping techniques.
4. The unsupervised cluster-based cross-domain pedestrian re-identification method according to claim 1, wherein the final feature representation of the image in step (3) is obtained by stitching a global feature vector and a local feature vector.
5. The unsupervised cluster-based cross-domain pedestrian re-identification method according to claim 1, wherein the step (4) is implemented as follows:
calculating the distance between the characteristic vectors by adopting Euclidean distance, and defining a label as y for a certain image0And according to the distance between the residual unmarked pictures and the images of the defined pseudo labels and a self-defined threshold value, if the distance between a certain image and the defined pseudo labels is greater than the threshold value, marking the image as the pseudo label, otherwise, defaulting that the image and a certain picture in the defined pseudo labels belong to the same pedestrian.
6. The unsupervised cluster-based cross-domain pedestrian re-identification method according to claim 1, wherein the loss function in step (6) is as follows:
and randomly extracting P identities and K instances of each small batch to meet the requirement of loss of the batch-hard triple:
the method comprises the following steps that a hardest positive in a triple is selected as a sample which is farthest away from an anchor in the same class, and the hardest positive is selected as a sample which is closest to the anchor in a non-similar class;
the loss function for camera invariance is:
wherein n istIs the number of real target pictures in the training batch,
is the number of samples generated by the camera style migration,
c is the number of cameras; the overall loss function is:
where β ∈ (0,1) is the weight of the loss of camera invariance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911172875.9A CN110942025A (en) | 2019-11-26 | 2019-11-26 | Unsupervised cross-domain pedestrian re-identification method based on clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911172875.9A CN110942025A (en) | 2019-11-26 | 2019-11-26 | Unsupervised cross-domain pedestrian re-identification method based on clustering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110942025A true CN110942025A (en) | 2020-03-31 |
Family
ID=69908278
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911172875.9A Withdrawn CN110942025A (en) | 2019-11-26 | 2019-11-26 | Unsupervised cross-domain pedestrian re-identification method based on clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110942025A (en) |
Cited By (44)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111476168A (en) * | 2020-04-08 | 2020-07-31 | 山东师范大学 | Cross-domain pedestrian re-identification method and system based on three stages |
| CN111582185A (en) * | 2020-05-11 | 2020-08-25 | 北京百度网讯科技有限公司 | Method and apparatus for recognizing image |
| CN111597887A (en) * | 2020-04-08 | 2020-08-28 | 北京大学 | Pedestrian re-identification method and system |
| CN111598004A (en) * | 2020-05-18 | 2020-08-28 | 北京星闪世图科技有限公司 | Progressive-enhancement self-learning unsupervised cross-domain pedestrian re-identification method |
| CN111639540A (en) * | 2020-04-30 | 2020-09-08 | 中国海洋大学 | Semi-supervised character re-recognition method based on camera style and human body posture adaptation |
| CN111738172A (en) * | 2020-06-24 | 2020-10-02 | 中国科学院自动化研究所 | Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering |
| CN111860678A (en) * | 2020-07-29 | 2020-10-30 | 中国矿业大学 | Unsupervised cross-domain pedestrian re-identification method based on clustering |
| CN111881714A (en) * | 2020-05-22 | 2020-11-03 | 北京交通大学 | Unsupervised cross-domain pedestrian re-identification method |
| CN111914668A (en) * | 2020-07-08 | 2020-11-10 | 浙江大华技术股份有限公司 | Pedestrian re-identification method, device and system based on image enhancement technology |
| CN111931655A (en) * | 2020-08-11 | 2020-11-13 | 山东建筑大学 | Unsupervised video pedestrian re-identification method and system based on local-global cooperation |
| CN111931619A (en) * | 2020-07-31 | 2020-11-13 | 杭州电子科技大学 | Cross-domain gait re-identification method based on optimized clustering algorithm |
| CN111967294A (en) * | 2020-06-23 | 2020-11-20 | 南昌大学 | Unsupervised domain self-adaptive pedestrian re-identification method |
| CN112016402A (en) * | 2020-08-04 | 2020-12-01 | 杰创智能科技股份有限公司 | Unsupervised learning-based pedestrian re-identification field self-adaption method and device |
| CN112016687A (en) * | 2020-08-20 | 2020-12-01 | 浙江大学 | Cross-domain pedestrian re-identification method based on complementary pseudo label |
| CN112036322A (en) * | 2020-09-01 | 2020-12-04 | 清华大学 | Method, system and device for constructing cross-domain pedestrian re-identification model of multi-task network |
| CN112069940A (en) * | 2020-08-24 | 2020-12-11 | 武汉大学 | Cross-domain pedestrian re-identification method based on staged feature learning |
| CN112069920A (en) * | 2020-08-18 | 2020-12-11 | 武汉大学 | Cross-domain pedestrian re-identification method based on attribute feature driven clustering |
| CN112232439A (en) * | 2020-11-06 | 2021-01-15 | 四川云从天府人工智能科技有限公司 | Method and system for updating pseudo label in unsupervised ReID |
| CN112287959A (en) * | 2020-06-01 | 2021-01-29 | 北京沃东天骏信息技术有限公司 | Training method of target re-recognition model and target re-recognition method and device |
| CN112395997A (en) * | 2020-11-19 | 2021-02-23 | 中山大学 | Weak supervision training method of pedestrian re-recognition model based on micrographic learning |
| CN112418331A (en) * | 2020-11-26 | 2021-02-26 | 国网甘肃省电力公司电力科学研究院 | Clustering fusion-based semi-supervised learning pseudo label assignment method |
| CN112633071A (en) * | 2020-11-30 | 2021-04-09 | 之江实验室 | Pedestrian re-identification data domain adaptation method based on data style decoupling content migration |
| CN112686135A (en) * | 2020-12-29 | 2021-04-20 | 中南大学 | Generalized pedestrian re-identification method based on distribution fitting |
| CN112686200A (en) * | 2021-01-11 | 2021-04-20 | 中山大学 | Pedestrian re-identification method and system based on multi-scheme parallel attention mechanism |
| CN112733695A (en) * | 2021-01-04 | 2021-04-30 | 电子科技大学 | Unsupervised key frame selection method in pedestrian re-identification field |
| CN112766218A (en) * | 2021-01-30 | 2021-05-07 | 上海工程技术大学 | Cross-domain pedestrian re-identification method and device based on asymmetric joint teaching network |
| CN112784674A (en) * | 2020-11-13 | 2021-05-11 | 北京航空航天大学 | Cross-domain identification method of key personnel search system based on class center self-adaption |
| CN112800967A (en) * | 2021-01-29 | 2021-05-14 | 重庆邮电大学 | Posture-driven shielded pedestrian re-recognition method |
| CN112801019A (en) * | 2021-02-09 | 2021-05-14 | 福州大学 | Method and system for eliminating re-identification deviation of unsupervised vehicle based on synthetic data |
| CN112836675A (en) * | 2021-03-01 | 2021-05-25 | 中山大学 | Unsupervised pedestrian re-identification method and system based on clustering-generated pseudo label |
| CN112966647A (en) * | 2021-03-25 | 2021-06-15 | 东北林业大学 | Pedestrian re-identification method based on layer-by-layer clustering and enhanced discrimination |
| CN113076984A (en) * | 2021-03-29 | 2021-07-06 | 海南智晶科技有限公司 | Cross-domain pedestrian re-identification method based on median clustering and global classification |
| CN113378632A (en) * | 2021-04-28 | 2021-09-10 | 南京大学 | Unsupervised domain pedestrian re-identification algorithm based on pseudo label optimization |
| CN113553970A (en) * | 2021-07-29 | 2021-10-26 | 广联达科技股份有限公司 | Pedestrian re-identification method, device, equipment and readable storage medium |
| CN113627380A (en) * | 2021-08-20 | 2021-11-09 | 中国矿业大学 | Cross-vision-field pedestrian re-identification method and system for intelligent security and early warning |
| CN113822262A (en) * | 2021-11-25 | 2021-12-21 | 之江实验室 | Pedestrian re-identification method based on unsupervised learning |
| CN114067356A (en) * | 2021-10-21 | 2022-02-18 | 电子科技大学 | Pedestrian re-identification method based on joint local guidance and attribute clustering |
| CN114140826A (en) * | 2021-12-03 | 2022-03-04 | 北京交通大学 | Target re-identification method based on camera feature separation |
| CN114882521A (en) * | 2022-03-30 | 2022-08-09 | 河北工业大学 | Unsupervised pedestrian re-identification method and unsupervised pedestrian re-identification device based on multi-branch network |
| CN116089652A (en) * | 2023-04-07 | 2023-05-09 | 中国科学院自动化研究所 | Unsupervised training method and device of visual retrieval model and electronic equipment |
| CN116258751A (en) * | 2023-05-08 | 2023-06-13 | 中国人民解放军国防科技大学 | Security patrol management method, system and equipment based on unsupervised learning |
| CN116763259A (en) * | 2023-08-17 | 2023-09-19 | 普希斯(广州)科技股份有限公司 | Multi-dimensional control method and device for beauty equipment and beauty equipment |
| CN117612206A (en) * | 2023-11-27 | 2024-02-27 | 深圳市大数据研究院 | Pedestrian re-recognition network model generation method, device, computer equipment and medium |
| CN117636100A (en) * | 2024-01-25 | 2024-03-01 | 北京航空航天大学杭州创新研究院 | Pre-training task model adjustment processing method and device, electronic equipment and medium |
-
2019
- 2019-11-26 CN CN201911172875.9A patent/CN110942025A/en not_active Withdrawn
Cited By (72)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111597887A (en) * | 2020-04-08 | 2020-08-28 | 北京大学 | Pedestrian re-identification method and system |
| CN111476168A (en) * | 2020-04-08 | 2020-07-31 | 山东师范大学 | Cross-domain pedestrian re-identification method and system based on three stages |
| CN111476168B (en) * | 2020-04-08 | 2022-06-21 | 山东师范大学 | Cross-domain pedestrian re-identification method and system based on three stages |
| CN111597887B (en) * | 2020-04-08 | 2023-02-03 | 北京大学 | Pedestrian re-identification method and system |
| CN111639540B (en) * | 2020-04-30 | 2023-08-04 | 中国海洋大学 | Semi-supervised character re-recognition method based on camera style and human body posture adaptation |
| CN111639540A (en) * | 2020-04-30 | 2020-09-08 | 中国海洋大学 | Semi-supervised character re-recognition method based on camera style and human body posture adaptation |
| CN111582185A (en) * | 2020-05-11 | 2020-08-25 | 北京百度网讯科技有限公司 | Method and apparatus for recognizing image |
| CN111582185B (en) * | 2020-05-11 | 2023-06-30 | 北京百度网讯科技有限公司 | Method and device for recognizing images |
| CN111598004A (en) * | 2020-05-18 | 2020-08-28 | 北京星闪世图科技有限公司 | Progressive-enhancement self-learning unsupervised cross-domain pedestrian re-identification method |
| CN111598004B (en) * | 2020-05-18 | 2023-12-08 | 江苏星闪世图科技(集团)有限公司 | Progressive reinforcement self-learning unsupervised cross-domain pedestrian re-identification method |
| CN111881714A (en) * | 2020-05-22 | 2020-11-03 | 北京交通大学 | Unsupervised cross-domain pedestrian re-identification method |
| CN111881714B (en) * | 2020-05-22 | 2023-11-21 | 北京交通大学 | Unsupervised cross-domain pedestrian re-identification method |
| CN112287959A (en) * | 2020-06-01 | 2021-01-29 | 北京沃东天骏信息技术有限公司 | Training method of target re-recognition model and target re-recognition method and device |
| CN111967294A (en) * | 2020-06-23 | 2020-11-20 | 南昌大学 | Unsupervised domain self-adaptive pedestrian re-identification method |
| CN111967294B (en) * | 2020-06-23 | 2022-05-20 | 南昌大学 | Unsupervised domain self-adaptive pedestrian re-identification method |
| CN111738172B (en) * | 2020-06-24 | 2021-02-12 | 中国科学院自动化研究所 | Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering |
| CN111738172A (en) * | 2020-06-24 | 2020-10-02 | 中国科学院自动化研究所 | Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering |
| CN111914668A (en) * | 2020-07-08 | 2020-11-10 | 浙江大华技术股份有限公司 | Pedestrian re-identification method, device and system based on image enhancement technology |
| CN111860678A (en) * | 2020-07-29 | 2020-10-30 | 中国矿业大学 | Unsupervised cross-domain pedestrian re-identification method based on clustering |
| CN111860678B (en) * | 2020-07-29 | 2024-02-27 | 中国矿业大学 | Unsupervised cross-domain pedestrian re-identification method based on clustering |
| CN111931619A (en) * | 2020-07-31 | 2020-11-13 | 杭州电子科技大学 | Cross-domain gait re-identification method based on optimized clustering algorithm |
| CN111931619B (en) * | 2020-07-31 | 2024-04-05 | 杭州电子科技大学 | Cross-domain gait re-identification method based on optimization clustering algorithm |
| CN112016402B (en) * | 2020-08-04 | 2024-05-17 | 杰创智能科技股份有限公司 | Self-adaptive method and device for pedestrian re-recognition field based on unsupervised learning |
| CN112016402A (en) * | 2020-08-04 | 2020-12-01 | 杰创智能科技股份有限公司 | Unsupervised learning-based pedestrian re-identification field self-adaption method and device |
| CN111931655A (en) * | 2020-08-11 | 2020-11-13 | 山东建筑大学 | Unsupervised video pedestrian re-identification method and system based on local-global cooperation |
| CN111931655B (en) * | 2020-08-11 | 2022-10-25 | 山东建筑大学 | Unsupervised video pedestrian re-identification method and system based on local-global cooperation |
| CN112069920A (en) * | 2020-08-18 | 2020-12-11 | 武汉大学 | Cross-domain pedestrian re-identification method based on attribute feature driven clustering |
| CN112016687A (en) * | 2020-08-20 | 2020-12-01 | 浙江大学 | Cross-domain pedestrian re-identification method based on complementary pseudo label |
| CN112016687B (en) * | 2020-08-20 | 2022-05-10 | 浙江大学 | Cross-domain pedestrian re-identification method based on complementary pseudo label |
| CN112069940A (en) * | 2020-08-24 | 2020-12-11 | 武汉大学 | Cross-domain pedestrian re-identification method based on staged feature learning |
| CN112036322A (en) * | 2020-09-01 | 2020-12-04 | 清华大学 | Method, system and device for constructing cross-domain pedestrian re-identification model of multi-task network |
| CN112036322B (en) * | 2020-09-01 | 2022-12-06 | 清华大学 | Method, system and device for constructing cross-domain pedestrian re-identification model of multi-task network |
| CN112232439A (en) * | 2020-11-06 | 2021-01-15 | 四川云从天府人工智能科技有限公司 | Method and system for updating pseudo label in unsupervised ReID |
| CN112232439B (en) * | 2020-11-06 | 2024-04-05 | 四川云从天府人工智能科技有限公司 | Pseudo tag updating method and system in unsupervised ReID |
| CN112784674A (en) * | 2020-11-13 | 2021-05-11 | 北京航空航天大学 | Cross-domain identification method of key personnel search system based on class center self-adaption |
| CN112395997A (en) * | 2020-11-19 | 2021-02-23 | 中山大学 | Weak supervision training method of pedestrian re-recognition model based on micrographic learning |
| CN112395997B (en) * | 2020-11-19 | 2023-11-24 | 中山大学 | Weak supervision training method based on pedestrian re-recognition model capable of micro-graph learning |
| CN112418331A (en) * | 2020-11-26 | 2021-02-26 | 国网甘肃省电力公司电力科学研究院 | Clustering fusion-based semi-supervised learning pseudo label assignment method |
| CN112633071A (en) * | 2020-11-30 | 2021-04-09 | 之江实验室 | Pedestrian re-identification data domain adaptation method based on data style decoupling content migration |
| CN112686135A (en) * | 2020-12-29 | 2021-04-20 | 中南大学 | Generalized pedestrian re-identification method based on distribution fitting |
| CN112733695A (en) * | 2021-01-04 | 2021-04-30 | 电子科技大学 | Unsupervised key frame selection method in pedestrian re-identification field |
| CN112733695B (en) * | 2021-01-04 | 2023-04-25 | 电子科技大学 | Unsupervised keyframe selection method in pedestrian re-identification field |
| CN112686200A (en) * | 2021-01-11 | 2021-04-20 | 中山大学 | Pedestrian re-identification method and system based on multi-scheme parallel attention mechanism |
| CN112800967B (en) * | 2021-01-29 | 2022-05-17 | 重庆邮电大学 | Posture-driven shielded pedestrian re-recognition method |
| CN112800967A (en) * | 2021-01-29 | 2021-05-14 | 重庆邮电大学 | Posture-driven shielded pedestrian re-recognition method |
| CN112766218A (en) * | 2021-01-30 | 2021-05-07 | 上海工程技术大学 | Cross-domain pedestrian re-identification method and device based on asymmetric joint teaching network |
| CN112801019B (en) * | 2021-02-09 | 2022-07-08 | 福州大学 | Method and system for eliminating re-identification deviation of unsupervised vehicle based on synthetic data |
| CN112801019A (en) * | 2021-02-09 | 2021-05-14 | 福州大学 | Method and system for eliminating re-identification deviation of unsupervised vehicle based on synthetic data |
| CN112836675B (en) * | 2021-03-01 | 2023-06-23 | 中山大学 | Unsupervised pedestrian re-identification method and system for generating pseudo tags based on clusters |
| CN112836675A (en) * | 2021-03-01 | 2021-05-25 | 中山大学 | Unsupervised pedestrian re-identification method and system based on clustering-generated pseudo label |
| CN112966647A (en) * | 2021-03-25 | 2021-06-15 | 东北林业大学 | Pedestrian re-identification method based on layer-by-layer clustering and enhanced discrimination |
| CN113076984A (en) * | 2021-03-29 | 2021-07-06 | 海南智晶科技有限公司 | Cross-domain pedestrian re-identification method based on median clustering and global classification |
| CN113378632A (en) * | 2021-04-28 | 2021-09-10 | 南京大学 | Unsupervised domain pedestrian re-identification algorithm based on pseudo label optimization |
| CN113378632B (en) * | 2021-04-28 | 2024-04-12 | 南京大学 | Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method |
| CN113553970A (en) * | 2021-07-29 | 2021-10-26 | 广联达科技股份有限公司 | Pedestrian re-identification method, device, equipment and readable storage medium |
| CN113627380B (en) * | 2021-08-20 | 2024-03-15 | 中国矿业大学 | Cross-vision pedestrian re-identification method and system for intelligent security and early warning |
| CN113627380A (en) * | 2021-08-20 | 2021-11-09 | 中国矿业大学 | Cross-vision-field pedestrian re-identification method and system for intelligent security and early warning |
| CN114067356A (en) * | 2021-10-21 | 2022-02-18 | 电子科技大学 | Pedestrian re-identification method based on joint local guidance and attribute clustering |
| CN114067356B (en) * | 2021-10-21 | 2023-05-09 | 电子科技大学 | Pedestrian re-recognition method based on combined local guidance and attribute clustering |
| CN113822262A (en) * | 2021-11-25 | 2021-12-21 | 之江实验室 | Pedestrian re-identification method based on unsupervised learning |
| CN113822262B (en) * | 2021-11-25 | 2022-04-15 | 之江实验室 | Pedestrian re-identification method based on unsupervised learning |
| CN114140826B (en) * | 2021-12-03 | 2024-05-31 | 北京交通大学 | Target re-identification method based on camera feature separation |
| CN114140826A (en) * | 2021-12-03 | 2022-03-04 | 北京交通大学 | Target re-identification method based on camera feature separation |
| CN114882521B (en) * | 2022-03-30 | 2024-09-13 | 河北工业大学 | Unsupervised pedestrian re-identification method and device based on multi-branch network |
| CN114882521A (en) * | 2022-03-30 | 2022-08-09 | 河北工业大学 | Unsupervised pedestrian re-identification method and unsupervised pedestrian re-identification device based on multi-branch network |
| CN116089652A (en) * | 2023-04-07 | 2023-05-09 | 中国科学院自动化研究所 | Unsupervised training method and device of visual retrieval model and electronic equipment |
| CN116258751A (en) * | 2023-05-08 | 2023-06-13 | 中国人民解放军国防科技大学 | Security patrol management method, system and equipment based on unsupervised learning |
| CN116763259A (en) * | 2023-08-17 | 2023-09-19 | 普希斯(广州)科技股份有限公司 | Multi-dimensional control method and device for beauty equipment and beauty equipment |
| CN116763259B (en) * | 2023-08-17 | 2023-12-08 | 普希斯(广州)科技股份有限公司 | Multi-dimensional control method and device for beauty equipment and beauty equipment |
| CN117612206A (en) * | 2023-11-27 | 2024-02-27 | 深圳市大数据研究院 | Pedestrian re-recognition network model generation method, device, computer equipment and medium |
| CN117636100B (en) * | 2024-01-25 | 2024-04-30 | 北京航空航天大学杭州创新研究院 | Pre-training task model adjustment processing method and device, electronic equipment and medium |
| CN117636100A (en) * | 2024-01-25 | 2024-03-01 | 北京航空航天大学杭州创新研究院 | Pre-training task model adjustment processing method and device, electronic equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110942025A (en) | Unsupervised cross-domain pedestrian re-identification method based on clustering | |
| CN110414462B (en) | Unsupervised cross-domain pedestrian re-identification method and system | |
| Abbas et al. | A comprehensive review of recent advances on deep vision systems | |
| Chen et al. | Underwater object detection using Invert Multi-Class Adaboost with deep learning | |
| CN111860678B (en) | Unsupervised cross-domain pedestrian re-identification method based on clustering | |
| CN112069920B (en) | Cross-domain pedestrian re-identification method based on attribute feature driven clustering | |
| Chuang et al. | A feature learning and object recognition framework for underwater fish images | |
| Kim et al. | Recurrent transformer networks for semantic correspondence | |
| Lin et al. | RSCM: Region selection and concurrency model for multi-class weather recognition | |
| Luo et al. | Hierarchical face parsing via deep learning | |
| Zhang et al. | Image-to-video person re-identification with temporally memorized similarity learning | |
| Liu et al. | Learning deep features via congenerous cosine loss for person recognition | |
| CN111598004B (en) | Progressive reinforcement self-learning unsupervised cross-domain pedestrian re-identification method | |
| CN109635708B (en) | Unsupervised pedestrian re-identification method based on three-data-set cross migration learning | |
| CN114882521B (en) | Unsupervised pedestrian re-identification method and device based on multi-branch network | |
| CN112668544B (en) | Pedestrian re-identification method based on hard sample confusion and feature robustness enhancement | |
| CN112784728A (en) | Multi-granularity clothes changing pedestrian re-identification method based on clothing desensitization network | |
| Wei et al. | Person re-identification based on deep learning—An overview | |
| Shi et al. | Instance enhancing loss: Deep identity-sensitive feature embedding for person search | |
| Wu et al. | An end-to-end exemplar association for unsupervised person re-identification | |
| Xia et al. | Face occlusion detection using deep convolutional neural networks | |
| Kim et al. | A shape preserving approach for salient object detection using convolutional neural networks | |
| Lee et al. | Property-specific aesthetic assessment with unsupervised aesthetic property discovery | |
| Zhang et al. | Category modeling from just a single labeling: Use depth information to guide the learning of 2d models | |
| Wang et al. | On branded handbag recognition |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20200331 |
|
| WW01 | Invention patent application withdrawn after publication |