CACENET.md

Devil's in the Details: Aligning Visual Clues for Conditional Embedding in Person Re-Identification

Devil's in the Details: Aligning Visual Clues for Conditional Embedding in Person Re-Identification is submitted to CVPR2021. In this paper, we proposes a strategy that integrates both visual clue alignment and conditional feature embedding into a unified ReID framework. Instead of using a pre-defined Adjacency Matrix, our CACE-Net uses a novel correspondence attention module where the visual clues is automatically predicted and dynamically adjusted during training

Here is our another tensorflow implementation CACENET.

peformance

config

yaml: 'experiment/cacenet/cacenet.yaml'

---

	Market1501 mAP&rank-1	DukeMTMC mAP&rank-1	download
paper	90.3/95.96	81.29/90.89	-
this implement	89.95/96.02	-	weight log

Citation

If you find this code useful, please cite the following paper:

@article{yu2020devil,
  title={Devil's in the Details: Aligning Visual Clues for Conditional Embedding in Person Re-Identification},
  author={Yu, Fufu and Jiang, Xinyang and Gong, Yifei and Zhao, Shizhen and Guo, Xiaowei and Zheng, Wei-Shi and Zheng, Feng and Sun, Xing},
  journal={arXiv e-prints},
  pages={arXiv--2009},
  year={2020}
}