Omni-Scale Feature Learning for Person Re-Identification

By Kaiyang Zhou, Yongxin Yang, Andrea Cavallaro, Tao A. Xiang. IEEE International Conference on Computer Vision, 2019

1,032 citations recorded by the source corpus.

As an instance-level recognition problem, person re-identification (ReID) relies on discriminative features, which not only capture different spatial scales but also encapsulate an arbitrary combination of multiple scales. We callse features of both homogeneous and heterogeneous scales omni-scale features. In this paper, a novel deep ReID CNN is designed, termed Omni-Scale Network (OSNet), for omni-scale feature learning. This is achieved by designing a residual block composed of multiple convolutional feature streams, each detecting features at a certain scale. Importantly, a novel unified aggregation gate is introduced to dynamically fuse multi-scale features with input-dependent channel-wise weights. To efficiently learn spatial-channel correlations and avoid overfitting, the building block uses both pointwise and depthwise convolutions. By stacking such blocks layer-by-layer, our OSNet is extremely lightweight and can be trained from scratch on existing ReID benchmarks. Despite its small model size, our OSNet achieves state-of-the-art performance on six person-ReID datasets. Code and models are available at: https://github.com/KaiyangZhou/deep-person-reid.

Subfields: Biomedical Engineering, Computer Vision and Pattern Recognition, Ocean Engineering

View Paper: https://www.semanticscholar.org/paper/45f4294a9ca31ca77a3393ba4e9499096d3eacc9

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. Tao A. Xiang
2. Rui Zhao
3. Cuiling Lan
4. Wenjun Zeng
5. Yu Qiao
6. Kaiyang Zhou
7. Li Zhang
8. Feng Zhu

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Xiaogang Wang
3. Ling Shao
4. Dacheng Tao
5. Yu Qiao
6. Tao A. Xiang
7. Yi Yang
8. Ming-Hsuan Yang

In Ocean Engineering

1. Yu Qiao
2. Tao A. Xiang
3. Kaiyang Zhou
4. Dapeng Chen
5. Rui Zhao
6. Yixiao Ge
7. Chen Change Loy
8. Steven C. H. Hoi

Citations in the corpus

• BoT-SORT: Robust Associations Multi-Pedestrian Tracking (2022, background): “Discriminating and re-identifying (ReID) objects by deep-appearance cues [43, 61, 28] has also become popular, but falls short in many cases, especially when scenes are crowded, due to partial occlusions of persons.”
• CLIP-ReID: Exploiting Vision-Language Model for Image Re-Identification without Concrete Text Labels (2022, background): “OSNet (Zhou et al. 2019) gives a lightweight model to deal with it.”
• Style Normalization and Restitution for Generalizable Person Re-Identification (2020, methodology): “Following [61, 56], we get each activation map by summarizing the feature maps along channels followed by a spatial l2 normalization.”
• Self-paced Contrastive Learning with Hybrid Memory for Domain Adaptive Object Re-ID (2020, methodology): “8 OSNet [64] ICCV’19 52.”
• Domain Generalization: A Survey (2021, methodology): “A common practice is to combine several face anti-spoofing datasets for model training and do evaluation on an unseen dataset, e.g., using CASIA-MFSD [79], Oulu-NPU [80] and MSU-MFSD [81] as the sources and Idiap Replay-Attack [82] as the target.”
• Deep Learning for Person Re-Identification: A Survey and Outlook (2020, methodology): “Modifying the network architecture to achieve a lightweight model is investigated in [79], [143], [144].”
• FastReID: A Pytorch Toolbox for General Instance Re-identification (2020, neutral): “Random erasing, Random patch [1] and Cutout [2] are also augmentation methods that randomly selects a rectangle region in an image and erases its pixels with random values, another image patch and zero values, making the model effectively reduce the risk of over-fitting and robust to occlusion.”
• Learning to Generate Novel Domains for Domain Generalization (2020, neutral): “For the CNN backbone, we employ the state-of-theart re-ID model, OSNet-IBN [62,61].”
• Counterfactual Attention Learning for Fine-Grained Visual Categorization and Re-identification (2021, neutral): “6 OSNet [68] 94.”
• Domain Generalization with MixStyle (2021, methodology): “We test MixStyle on two CNN architectures: ResNet-50 (He et al., 2016) and OSNet (Zhou et al., 2019).”