CN112906606B - Domain self-adaptive pedestrian re-identification method based on mutual divergence learning - Google Patents
Domain self-adaptive pedestrian re-identification method based on mutual divergence learning Download PDFInfo
- Publication number
- CN112906606B CN112906606B CN202110243888.1A CN202110243888A CN112906606B CN 112906606 B CN112906606 B CN 112906606B CN 202110243888 A CN202110243888 A CN 202110243888A CN 112906606 B CN112906606 B CN 112906606B
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- data set
- training
- samples
- learning
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 37
- 239000013598 vector Substances 0.000 claims abstract description 18
- 230000008485 antagonism Effects 0.000 claims abstract description 5
- 238000012163 sequencing technique Methods 0.000 claims abstract description 4
- 238000002156 mixing Methods 0.000 claims abstract description 3
- 239000000203 mixture Substances 0.000 claims abstract description 3
- 230000003044 adaptive effect Effects 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 2
- 230000008569 process Effects 0.000 description 5
- 230000006978 adaptation Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000002372 labelling Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
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
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- 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
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a domain self-adaptive pedestrian re-identification method based on mutual divergence learning, which comprises the following steps: preparing a pedestrian data set; pre-training is carried out on a source domain data set, and feature vectors of pictures are extracted on a target domain data set; clustering the images of the target domain data set based on density, and taking the serial numbers of the clusters as pseudo tags; adding outliers to the training samples using a strategy of antagonism; mixing the clustered samples and outliers, sending the mixture into a network, correcting the noise of the pseudo tag by adopting mutual divergence learning, inputting the pedestrian image to be queried into a trained pedestrian re-recognition model to obtain a pedestrian feature vector to be recognized, and comparing and sequencing the pedestrian feature vector with attribute features in a candidate library to obtain a pedestrian re-recognition result. The invention reduces the distribution difference between the source domain and the target domain, effectively utilizes the knowledge of the source domain, and finally the framework of the invention can learn the characteristics with robustness and authentication.
Description
Technical Field
The invention discloses a method for realizing self-adaptive pedestrian re-identification in an unsupervised domain by applying deep learning and mutual learning, and belongs to the field of computer vision.
Background
Pedestrian re-identification (re-ID) is a technique that aims to establish identity correspondence between different cameras, determine images of different cameras, or whether a particular pedestrian is present in a video sequence, and is generally considered as a sub-problem of image retrieval. Great attention has been paid to, and significant progress has been made in the last decade. The pedestrian re-identification technology is widely applied to tracking the track of a person in a larger area, and has high application value in the fields of robotics, intelligent video monitoring, automatic photo labeling and the like.
At present, compared with the mature face recognition technology, pedestrian re-recognition is still a difficult problem in the field of computer vision. The main challenge is that the same target is affected by visual angle change, illumination change, posture change, pedestrian shielding, background noise interference and the like under different cameras, so that characteristic representation under different visual angles has a certain degree of deviation. The existing pedestrian re-recognition method is generally based on deep learning, and basically comprises the steps of calculating the similarity or distance between images, sorting samples according to the similarity or distance, and finally finding out images belonging to the same person as the pedestrian image to be queried. However, due to the limitation of resolution of the monitoring video, it is difficult to directly search for the same target through the face in the application of the actual monitoring field, however, searching for pedestrians by using the external features such as the wearing appearance of the pedestrians is an alternative method.
Most of the work currently in studying pedestrian re-identification is focused on supervised learning methods, which rely heavily on the acquisition of large-scale data sets and accurate manual labeling, which is often a time-consuming and cumbersome task. While the re-ID method achieves very good results under the supervision of such large-scale data sets, they often suffer from catastrophic performance degradation when we apply the trained reid model directly to new camera systems due to the existence of domain differences. Thus, current research emphasis has been shifted to Unsupervised Domain Adaptation (UDA), which attempts to shift a model trained on a labeled source domain dataset to an unlabeled target domain dataset. Despite the remarkable progress made by the cluster-based UDA method, unavoidable pseudo tag noise due to limited transmissibility of source domain features, imperceptibility of target domain image tags, and imperfection of clustering results remains an obstacle factor for performance improvement.
Reference is made to:
[1].LIN,Yutian,et al.A bottom-up clustering approach to unsupervised person re-identification[C].Proceedings of the AAAI Conference on Artificial Intelligence,2019.p.8738-8745.
[2] wang, zhongdao, zheng, liang, li, yali, et al Linkage Based Face Clustering via Graph Convolution Network [ J ].
[3].Yixiao Ge,Dapeng Chen,and Hongsheng Li,“Mutual mean-teaching:Pseudo label refinery for unsupervised domain adaptation on person re-identification,”in International Conference on Learning Representations,2020.
[4].Yunpeng Zhai,Shijian Lu,Qixiang Ye,Xuebo Shan,Jie Chen,Rongrong Ji,and Yonghong Tian,“Ad-cluster:Augmented discriminative clustering for domain adaptive person re-identification,”in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,2020,pp.9021–9030.
Disclosure of Invention
In order to solve the defects of the scheme, the invention aims to provide a domain self-adaptive pedestrian re-identification method based on mutual divergence learning so as to reduce the distribution difference between a source domain and a target domain.
In order to achieve the above purpose, the invention adopts the following technical scheme:
a domain self-adaptive pedestrian re-identification method based on mutual divergence learning comprises the following steps:
step S1, preparing a pedestrian data set, wherein the pedestrian data set comprises a labeled source domain data set and an unlabeled target domain data set;
step S2, pre-training is carried out on a source domain data set, and feature vectors of pictures are extracted on a target domain data set;
step S3, clustering the images of the target domain data set based on density, and taking the serial numbers of the clusters as pseudo tags;
step S4, adding outliers into the training samples by using a strategy of antagonism;
and S5, mixing the clustered samples obtained in the step S3 and the outliers obtained in the step S4, sending the mixture into a network, correcting the noise of the pseudo tag by adopting mutual divergence learning, inputting the pedestrian image to be queried into a trained pedestrian re-recognition model to obtain a pedestrian feature vector to be recognized, and comparing and sequencing the feature vector with the attribute features in the candidate library to obtain a pedestrian re-recognition result.
In the step S2, a ResNet-50 model is utilized to conduct supervised pre-training on a source domain data set, then a pre-training model is utilized to initialize a training target domain data set, and characteristics of pictures in the target domain data set are extracted.
In the step S2, initializing by using a pre-trained model, and removing the last classification layer of the ResNet-50 model; extracting feature vectors of pictures on a target domain dataset is denoted as x= { X 1 ,x 2 ,...,x N N is the number of sample pictures in the dataset, each x is a 2048-dimensional feature vector.
In the step S2, the pre-trained model is optimized for cross entropy loss and triplet loss by using a loss function.
The step S3 includes:
step S31, calculating the distance between each image and all other images in the target domain data set;
step S32, grouping samples in a high-density region into clusters and keeping samples in a low-density region as outliers using a density-based clustering algorithm (DBSCAN) for the distance between each image;
step S33, for the images that have been clustered into clusters, using the numbers of the clusters in which they are located as their pseudo tags, and then performing optimization training on the pseudo tags.
In the step S4, an additional auxiliary network is used, and a part of clustered images are sent into the auxiliary network for learning, so that the auxiliary network can obtain the general characteristics of the clustered samples; the outliers are then fed into the auxiliary network, which can extract some outliers with small losses.
In the step S5, the training and the optimization of the pseudo tag are alternately performed by correcting the pseudo tag through the mutual divergence learning: in the early stage of training, the difference of two networks is maintained by inputting the same images but respectively carrying out random erasing, clipping and overturning and carrying out different parameter initialization; in the middle and later stages of training, adopting a divergence strategy to slow down the speed of two networks to reach consensus, and keeping continuous divergence between the two networks; executing each network to perform own prediction, and selecting samples with prediction differences between the two networks; based on these samples, each network further trains the divergent samples and updates their parameters.
In said step S5, a momentum-based moving average model is proposed for each peer-to-peer network.
Compared with the prior art, the technical scheme has the following beneficial effects:
(1) The invention provides an unsupervised domain self-adaptive pedestrian re-identification method based on mutual divergence learning, which reduces the distribution difference between a source domain and a target domain, effectively utilizes the knowledge of the source domain, and finally the framework of the invention can learn the characteristics with robustness and discrimination.
(2) The invention adopts the antagonism strategy to gradually add the outlier in the clustering result to the training process, thereby increasing the diversity and reliability of the training samples.
(3) Experimental results on three large datasets demonstrated the superiority of the framework presented in this invention over other up-to-date methods.
Drawings
Fig. 1 is a flow chart of domain adaptive pedestrian re-recognition based on mutual divergence learning.
Detailed Description
The present invention will be further explained below.
As shown in fig. 1, the domain adaptive pedestrian re-recognition method based on mutual divergence learning of the invention comprises the following steps:
step 1, preparation and pretreatment of a data set:
the dataset includes a source domain dataset with complete annotation information and a target domain dataset without any manual annotation information.
Three public data sets Market-1501, dukeMTMC-ReID and MSMT17 commonly used in the field of pedestrian re-recognition research are used as the data sets of the training model. Market-1501 this dataset contains 1501 pedestrians and 32688 tagged images from 6 different cameras. Of all images, 12936 images of 751 pedestrians were used for training, 3368 images of another 750 pedestrians were used for query, and 19732 images of 750 pedestrians were used as a library of bellery charts. The identities between the training image and the gallery image are disjoint. The DukeMTMC-ReID dataset consisted of video captured outdoors from 8 cameras containing 1404 pedestrians and 36411 tagged images, with a training set of 16522 images of 702 pedestrians, 17661 images for gallery and 2228 images for query. The MSMT17 dataset is the most challenging dataset taken by 15 cameras, containing 126441 images of 4101 pedestrians.
Step 2: the model ResNet-50 with the best feature extraction effect in the pedestrian re-recognition field is utilized to pretrain on the source domain data set and extract the features of the target domain sample. A supervised pre-training is performed on the source domain dataset with complete tag information, and then the training target domain dataset is initialized with a pre-trained model. Specifically, supervised training is performed on the source domain dataset Market1501, and the network is optimized using cross entropy loss and triplet loss, resulting in a pre-trained model. Then using the model as an initial model for target domain dataset learning, and removing the ResNet-50 last classification layer; sending the unlabeled source domain data set into a pre-trained model, and extracting the feature vector of each picture; the sample characteristics are expressed as x= { X 1 ,x 2 ,...,x N N is the number of sample pictures in the dataset, each x is a 2048-dimensional feature vector.
Step 3, for the feature vector of each image in step 2, calculating the distance between the images, then using a density-based clustering algorithm (DBSCAN) for the distance, grouping samples in a high density area into clusters, and keeping samples in a low density area as outliers, so that the target domain dataset can be divided into clustered points and outliers. For points that have been clustered, the number of the cluster in which they are located is used as their pseudo tag. For outliers, a strategy of resistance is used in step 4 to add small loss samples to the training process.
Step 4, for the generated outliers of step 3, an attempt is made to gradually append some of the small-loss samples in the outliers to the training process using an auxiliary network, which not only effectively uses these difficult samples, but also further improves the performance of the model. The primary model and secondary network are trained alternately so that small loss samples of relatively high confidence can be added smoothly to the training process. This results in a more reliable, more diverse training sample containing clustered samples and less missing outliers. However, the supervision information may be noisy due to imperfections in the clustering results and the addition of outliers.
In order to prevent the re-ID model from being affected by noisy labels, a mutual divergence learning is proposed to correct the pseudo labels, and training and optimization of the pseudo labels are alternated. In the early stage of training, the difference of two networks is maintained by inputting the same image but respectively performing random erasing, clipping and flipping, performing different parameter initialization and the like. In the middle and later stages of training, a bifurcation strategy is attempted to slow down the speed at which the two networks reach consensus and maintain a persistent bifurcation between the two networks. Specifically, each network is executed to make its own prediction, and a sample in which there is a prediction difference between two networks is selected. Based on these hard samples, which are in fact difficult but valuable, each network further trains these divergent samples and updates their parameters. In order to prevent the two networks from converging close to each other, a momentum-based moving average model is proposed for each peer-to-peer network.
And finally, inputting the pedestrian image to be queried into a trained pedestrian re-recognition model to obtain a pedestrian feature vector to be recognized, and comparing and sequencing the pedestrian feature vector with the attribute features in the candidate library to obtain a pedestrian re-recognition result.
In summary, according to the domain adaptive pedestrian re-recognition method based on mutual divergence learning, clustering is performed on an unlabeled dataset by using DBSCAN, and then outliers with small loss are added into a training process by using a strategy of antagonism, so that the pedestrian characteristics with identification are learned. The method reduces the complexity of calculation, saves a great amount of manual labeling cost, and effectively improves the accuracy of pedestrian matching retrieval.
The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention.
Claims (6)
1. A domain self-adaptive pedestrian re-identification method based on mutual divergence learning is characterized by comprising the following steps of: the method comprises the following steps:
step S1, preparing a pedestrian data set, wherein the pedestrian data set comprises a labeled source domain data set and an unlabeled target domain data set;
step S2, pre-training is carried out on a source domain data set, and feature vectors of pictures are extracted on a target domain data set;
step S3, clustering the images of the target domain data set based on density, and taking the serial numbers of the clusters as pseudo tags;
step S4, adding outliers into the training samples by using a strategy of antagonism;
in the step S4, an additional auxiliary network is used, and a part of clustered images are sent into the auxiliary network for learning, so that the auxiliary network can obtain the general characteristics of the clustered samples; then the outliers are sent into an auxiliary network, so that some outliers with small loss can be extracted;
step S5, mixing the clustered samples obtained in the step S3 and the outliers obtained in the step S4, sending the mixture into a network, correcting the noise of the pseudo tag by adopting mutual divergence learning, inputting the pedestrian image to be queried into a trained pedestrian re-recognition model to obtain a pedestrian feature vector to be recognized, and comparing and sequencing the pedestrian feature vector with the attribute features in the candidate library to obtain a pedestrian re-recognition result;
in the step S5, the training and the optimization of the pseudo tag are alternately performed by correcting the pseudo tag through the mutual divergence learning: in the early stage of training, the difference of two networks is maintained by inputting the same images but respectively carrying out random erasing, clipping and overturning and carrying out different parameter initialization; in the middle and later stages of training, adopting a divergence strategy to slow down the speed of two networks to reach consensus, and keeping continuous divergence between the two networks; executing each network to perform own prediction, and selecting samples with prediction differences between the two networks; based on these samples, each network further trains the divergent samples and updates their parameters.
2. The domain adaptive pedestrian re-recognition method based on the mutual divergence learning as claimed in claim 1, wherein: in the step S2, a ResNet-50 model is utilized to conduct supervised pre-training on a source domain data set, then a pre-training model is utilized to initialize a training target domain data set, and characteristics of pictures in the target domain data set are extracted.
3. The domain adaptive pedestrian re-recognition method based on the mutual divergence learning as claimed in claim 2, wherein: in the step S2, initializing by using a pre-trained model, and removing the last classification layer of the ResNet-50 model; extracting feature vectors of pictures on a target domain dataset is denoted as x= { X 1 ,x 2 ,...,x N N is the number of sample pictures in the dataset, each X is a 2048-dimensional feature vector.
4. The domain adaptive pedestrian re-recognition method based on the mutual divergence learning as claimed in claim 2, wherein: in the step S2, the pre-trained model is optimized for cross entropy loss and triplet loss by using a loss function.
5. The domain adaptive pedestrian re-recognition method based on the mutual divergence learning as claimed in claim 1, wherein: the step S3 includes:
step S31, calculating the distance between each image and all other images in the target domain data set;
step S32, grouping samples in a high-density area into clusters and keeping samples in a low-density area as outliers by using a density-based clustering algorithm for the distance between each image;
step S33, for the images that have been clustered into clusters, using the numbers of the clusters in which they are located as their pseudo tags, and then performing optimization training on the pseudo tags.
6. The domain adaptive pedestrian re-recognition method based on the mutual divergence learning as claimed in claim 1, wherein: in said step S5, a momentum-based moving average model is proposed for each peer-to-peer network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110243888.1A CN112906606B (en) | 2021-03-05 | 2021-03-05 | Domain self-adaptive pedestrian re-identification method based on mutual divergence learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110243888.1A CN112906606B (en) | 2021-03-05 | 2021-03-05 | Domain self-adaptive pedestrian re-identification method based on mutual divergence learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112906606A CN112906606A (en) | 2021-06-04 |
CN112906606B true CN112906606B (en) | 2024-04-02 |
Family
ID=76108309
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110243888.1A Active CN112906606B (en) | 2021-03-05 | 2021-03-05 | Domain self-adaptive pedestrian re-identification method based on mutual divergence learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112906606B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113255573B (en) * | 2021-06-17 | 2021-10-01 | 成都东方天呈智能科技有限公司 | Pedestrian re-identification method based on mixed cluster center label learning and storage medium |
CN113657254B (en) * | 2021-08-16 | 2023-09-19 | 浙江大学 | Pedestrian re-identification domain adaptation method based on reliable value sample and new identity sample mining |
CN114022904B (en) * | 2021-11-05 | 2023-01-24 | 湖南大学 | Noise robust pedestrian re-identification method based on two stages |
CN114187655B (en) * | 2021-11-29 | 2024-08-13 | 大连理工大学 | Unsupervised pedestrian re-recognition method based on joint training strategy |
CN114283287B (en) * | 2022-03-09 | 2022-05-06 | 南京航空航天大学 | Robust field adaptive image learning method based on self-training noise label correction |
CN115937617B (en) * | 2023-03-06 | 2023-05-30 | 支付宝(杭州)信息技术有限公司 | Risk identification model training and risk control method, device and equipment |
CN117612201B (en) * | 2023-10-20 | 2024-05-28 | 南通大学 | Single-sample pedestrian re-identification method based on feature compression |
Citations (2)
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 |
CN112347995A (en) * | 2020-11-30 | 2021-02-09 | 中国科学院自动化研究所 | Unsupervised pedestrian re-identification method based on fusion of pixel and feature transfer |
-
2021
- 2021-03-05 CN CN202110243888.1A patent/CN112906606B/en active Active
Patent Citations (2)
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 |
CN112347995A (en) * | 2020-11-30 | 2021-02-09 | 中国科学院自动化研究所 | Unsupervised pedestrian re-identification method based on fusion of pixel and feature transfer |
Also Published As
Publication number | Publication date |
---|---|
CN112906606A (en) | 2021-06-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112906606B (en) | Domain self-adaptive pedestrian re-identification method based on mutual divergence learning | |
Fu et al. | Self-similarity grouping: A simple unsupervised cross domain adaptation approach for person re-identification | |
CN110598554B (en) | Multi-person posture estimation method based on counterstudy | |
Kim et al. | Self-training and adversarial background regularization for unsupervised domain adaptive one-stage object detection | |
AU2021103625A4 (en) | Remote sensing image semantic segmentation method based on contrastive self-supervised learning | |
Spencer et al. | Defeat-net: General monocular depth via simultaneous unsupervised representation learning | |
Von Stumberg et al. | Gn-net: The gauss-newton loss for multi-weather relocalization | |
CN112101150B (en) | Multi-feature fusion pedestrian re-identification method based on orientation constraint | |
CN111832514B (en) | Unsupervised pedestrian re-identification method and unsupervised pedestrian re-identification device based on soft multiple labels | |
CN110942025A (en) | Unsupervised cross-domain pedestrian re-identification method based on clustering | |
CN109784258A (en) | A kind of pedestrian's recognition methods again cut and merged based on Analysis On Multi-scale Features | |
CN112819065B (en) | Unsupervised pedestrian sample mining method and unsupervised pedestrian sample mining system based on multi-clustering information | |
CN112651262B (en) | Cross-modal pedestrian re-identification method based on self-adaptive pedestrian alignment | |
CN111783521B (en) | Pedestrian re-identification method based on low-rank prior guidance and based on domain invariant information separation | |
CN111598004B (en) | Progressive reinforcement self-learning unsupervised cross-domain pedestrian re-identification method | |
CN113034545A (en) | Vehicle tracking method based on CenterNet multi-target tracking algorithm | |
CN110991321B (en) | Video pedestrian re-identification method based on tag correction and weighting feature fusion | |
CN112668544B (en) | Pedestrian re-identification method based on hard sample confusion and feature robustness enhancement | |
CN111209799A (en) | Pedestrian searching method based on partial shared network and cosine interval loss function | |
Wang et al. | Multiple pedestrian tracking with graph attention map on urban road scene | |
Wang et al. | Regression forest based RGB-D visual relocalization using coarse-to-fine strategy | |
Zeng et al. | Anchor association learning for unsupervised video person re-identification | |
CN109165565A (en) | A kind of video object discovery and dividing method based on Coupled Dynamic Markov Network | |
CN114580492B (en) | Cross-domain pedestrian re-recognition method based on mutual learning | |
CN114973099A (en) | Intelligent object searching method and system based on traceable target identification |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |