Deep Attributes Driven Multi-Camera Person Re-identification

By Chi Su, Shiliang Zhang, Junliang Xing, Wen Gao, Qi Tian. European Conference on Computer Vision, 2016

415 citations recorded by the source corpus.

The visual appearance of a person is easily affected by many factors like pose variations, viewpoint changes and camera parameter differences. This makes person Re-Identification (ReID) among multiple cameras a very challenging task. This work is motivated to learn mid-level human attributes which are robust to such visual appearance variations. And we propose a semi-supervised attribute learning framework which progressively boosts the accuracy of attributes only using a limited number of labeled data. Specifically, this framework involves a three-stage training. A deep Convolutional Neural Network (dCNN) is first trained on an independent dataset labeled with attributes. Then it is fine-tuned on another dataset only labeled with person IDs using our defined triplet loss. Finally, the updated dCNN predicts attribute labels for the target dataset, which is combined with the independent dataset for the final round of fine-tuning. The predicted attributes, namely \emph{deep attributes} exhibit superior generalization ability across different datasets. By directly using the deep attributes with simple Cosine distance, we have obtained surprisingly good accuracy on four person ReID datasets. Experiments also show that a simple metric learning modular further boosts our method, making it significantly outperform many recent works.

Subfields: Biomedical Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/2788f382e4396290acfc8b21df45cc811586e66e

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

1. Yi Yang
2. Liang Zheng
3. Zhedong Zheng
4. Qi Tian
5. Shiliang Zhang
6. Wen Gao
7. Tao A. Xiang
8. Shengjin Wang

In Computer Vision and Pattern Recognition

1. Yi Yang
2. Luc Van Gool
3. Dacheng Tao
4. Xiaogang Wang
5. Liang Zheng
6. Qi Tian
7. Ming-Hsuan Yang
8. Ling Shao

Citations in the corpus

• Person Re-identification: Past, Present and Future (2016, methodology): “In [50], [75], attribute annotations from independent datasets are employed to represent the re-ID images.”
• Scalable Person Re-identification on Supervised Smoothed Manifold (2017, background): “Thanks to plenty of training images provided, training deep neural networks becomes feasible on this dataset and preferred by most previous works [47, 50, 49].”
• Person Re-identification by Deep Learning Multi-scale Representations (2017, neutral): “4 SSDAL [51] 39.”
• Beyond Triplet Loss: A Deep Quadruplet Network for Person Re-identification (2017, methodology): “In addition to a triplet loss, some deep learning methods [19, 1, 41, 35, 29, 32] address the person ReID problem from the classification aspect and adopt a binary classification loss to train their models.”
• Multiple object tracking: A literature review (2014, background): “Possible combinations include object segmentation [174], human re-identification [175], human pose estimation [17], and action recognition [18].”
• Attribute-Aware Attention Model for Fine-grained Representation Learning (2018, neutral): “[33] use attributes triplet loss to fine-tune CNNs.”
• Pose-Driven Deep Convolutional Model for Person Re-identification (2017, methodology): “More specifically, existing deep learning based person ReID approaches can be summarized into two categories: 1) use Softmax Loss with person ID labels to learn a global representation [1, 10, 53, 46, 42, 61, 13], and 2) first learn local representations using predefined rigid body parts, then fuse ”
• Spindle Net: Person Re-identification with Human Body Region Guided Feature Decomposition and Fusion (2017, methodology): “The proposed Spindle Net is compared with several state-of-the-art ReID approaches in recent two years, including the metric learning method WARCA-χ2 [11], the null space semi-supervised learning method NFST [32], the deep neural network PersonNet [28], the Siamese CNN with learnable gate S-CNN [25]”
• In Defense of the Triplet Loss for Person Re-Identification (2017, neutral): “However, the recently most suc- ∗Equal contribution.”
• GLAD: Global-Local-Alignment Descriptor for Pedestrian Retrieval (2017, background): “Person Re-ID can be tackled by classifying those gallery images [25, 36, 37, 45], or by an image retrieval procedure, i.”
• CVM-Net: Cross-View Matching Network for Image-Based Ground-to-Aerial Geo-Localization (2018, background): “The most widely used loss function in image retrieval task is the max-margin triplet loss [9, 3, 8, 11, 24, 42, 47, 14, 32, 19] that enforces the distances of positive pairs to be less than the distances of negative pairs.”