Robust Anchor Embedding for Unsupervised Video Person re-IDentification in the Wild

@inproceedings{Ye2018RobustAE,
  title={Robust Anchor Embedding for Unsupervised Video Person re-IDentification in the Wild},
  author={Mang Ye and Xiangyuan Lan and Pong Chi Yuen},
  booktitle={ECCV},
  year={2018}
}
This paper addresses the scalability and robustness issues of estimating labels from imbalanced unlabeled data for unsupervised video-based person re-identification (re-ID). To achieve it, we propose a novel Robust AnChor Embedding (RACE) framework via deep feature representation learning for large-scale unsupervised video re-ID. Within this framework, anchor sequences representing different persons are firstly selected to formulate an anchor graph which also initializes the CNN model to get… 

Dynamic Graph Co-Matching for Unsupervised Video-Based Person Re-Identification

TLDR
A dynamic graph matching (DGM) framework is designed, which improves the label estimation process by iteratively refining the graph structure with better similarity measurement learned from the intermediate estimated labels, which enhances the robustness against inaccurate matching output and noisy initial training data.

Multi-View Label Prediction for Unsupervised Learning Person Re-Identification

TLDR
This work proposes to learn pseudo-patch-labels, which brings two advantages: (1) Patch naturally alleviates the effect of backgrounds, occlusions, and carryings since they usually occupy small parts in images, thus overcome noisy labels, and (2) It is plausible that patches from different pedestrians belong to the same pseudo-identity-label.

Deep Graph Metric Learning for Weakly Supervised Person Re-Identification

TLDR
A weakly supervised person re-id modeling that aims to find the raw video clips where a given target person appears is considered and deep graph metric learning (DGML) is developed, which utilizes multiple weak video-level labels to learn discriminative features by means of a weak identity loss and a cross-video alignment loss.

Deep Active Learning for Video-based Person Re-identification

TLDR
A sampling criterion to choose the most TP-likely tracklet-pairs for annotation is proposed and Experimental results show that less than 3\% pairwise annotations are needed for the method to reach comparable performance with the fully-supervised setting.

Pose-Transformation and Radial Distance Clustering for Unsupervised Person Re-identification

TLDR
This work proposes an unsupervised approach to the person re-ID setup that enhances the discriminating ability of the learned features via a novel two-stage training strategy and introduces a novel radial distance loss, that attends to the fundamental as-pects of feature learning.

Video-Based Person Re-Identification Using Unsupervised Tracklet Matching

TLDR
This work proposes a novel algorithm for unsupervised video-based person re-id applications that allows the rejection of poor and noisy clusters while retaining the most discriminative ones for matching.

Iterative Local-Global Collaboration Learning Towards One-Shot Video Person Re-Identification

TLDR
An iterative local-global collaboration learning approach to learn robust and discriminative person representations that jointly considers the global video information and local frame sequence information to better capture the diverse appearance of the person for feature learning and pseudo-label estimation.

Learning to Align Multi-Camera Domains using Part-Aware Clustering for Unsupervised Video Person Re-Identification.

TLDR
This paper addresses the scalability issue by presenting deep representation learning without ID information across multiple cameras, and proposes a part-aware adaptation module, which effectively performs multi-camera domain invariant feature learning in different spatial regions.

Unsupervised Person Re-identification via Cross-Camera Similarity Exploration

TLDR
This paper proposes an iterative framework which overcomes the camera variance and achieves across-camera similarity exploration and demonstrates that the algorithm is not only superior to state-of-the-art unsupervised re-ID approaches, but also performs favorably compared with other competing un supervised domain adaptation methods (UDA) and semi-supervised learning methods.

Unsupervised Domain Adaptation in Person re-ID via k-Reciprocal Clustering and Large-Scale Heterogeneous Environment Synthesis

TLDR
Experimental results show that the proposed ktCUDA and SHRED approach achieves an average improvement of +5.7 mAP in re-identification performance when compared to existing state-of-the-art methods, as well as demonstrate better adaptability to different types of environments.
...

References

SHOWING 1-10 OF 59 REFERENCES

Dynamic Label Graph Matching for Unsupervised Video Re-identification

TLDR
A dynamic graph matching (DGM) method is proposed that iteratively updates the image graph and the label estimation process by learning a better feature space with intermediate estimated labels and is robust to noisy initial training data.

Semi-supervised Region Metric Learning for Person Re-identification

TLDR
A novel semi-supervised region metric learning method to improve person re-identification performance under imbalanced unlabeled data is proposed, which proposes to estimate positive neighbors by label propagation with cross person score distribution alignment.

Person Re-identification by Unsupervised ` 1 Graph Learning

TLDR
This work proposes a novel unsupervised Re-ID approach which requires no labelled training data yet is able to capture discriminative information for cross-view identity matching, and is based on a new graph regularised dictionary learning algorithm.

Stepwise Metric Promotion for Unsupervised Video Person Re-identification

TLDR
This paper proposes a stepwise metric promotion approach to estimate the identities of training tracklets, which iterates between cross-camera tracklet association and feature learning, and can eliminate the hard negative label matches.

Person Re-Identification by Unsupervised \ell _1 ℓ 1 Graph Learning

TLDR
This work proposes a novel unsupervised Re-ID approach which requires no labelled training data yet is able to capture discriminative information for cross-view identity matching, and is based on a new graph regularised dictionary learning algorithm.

Unsupervised Person Re-identification

TLDR
A progressive unsupervised learning (PUL) method to transfer pretrained deep representations to unseen domains and demonstrates that PUL outputs discriminative features that improve the re-ID accuracy.

Exploit the Unknown Gradually: One-Shot Video-Based Person Re-identification by Stepwise Learning

TLDR
This paper proposes an approach to exploiting unlabeled tracklets by gradually but steadily improving the discriminative capability of the Convolutional Neural Network feature representation via stepwise learning.

Scalable Person Re-identification on Supervised Smoothed Manifold

  • S. BaiX. BaiQ. Tian
  • Computer Science
    2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
  • 2017
TLDR
An unconventional manifold-preserving algorithm is proposed, which can make best use of supervision from training data, whose label information is given as pairwise constraints, scale up to large repositories with low on-line time complexity, and be plunged into most existing algorithms, serving as a generic postprocessing procedure to further boost the identification accuracies.

Unsupervised Salience Learning for Person Re-identification

TLDR
A novel perspective for person re-identification based on unsupervised salience learning, which applies adjacency constrained patch matching to build dense correspondence between image pairs, which shows effectiveness in handling misalignment caused by large viewpoint and pose variations.

Diversity Regularized Spatiotemporal Attention for Video-Based Person Re-identification

TLDR
A new spatiotemporal attention model is proposed that automatically discovers a diverse set of distinctive body parts in video clips of people across non-overlapping cameras and outperforms the state-of-the-art approaches by large margins on multiple metrics.
...