[공지사항]을 빙자한 안부와 근황 
Show more

Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

KeyRe-ID: Keypoint-Guided Person Re-Identification using Part-Aware Representation in Videos

Created by
  • Haebom

Author

Jinseong Kim, Jeonghoon Song, Gyeongseon Baek, Byeongjoon Noh

Outline

KeyRe-ID is a video-based person re-identification framework leveraging keypoints, which performs enhanced spatiotemporal representation learning via global and local branches. The global branch captures the overall identity semantics via Transformer-based temporal aggregation, while the local branch dynamically segments body regions based on keypoints to generate fine-grained part recognition features. Extensive experiments on MARS and iLIDS-VID benchmarks demonstrate state-of-the-art performance, achieving 91.73% mAP and 97.32% Rank-1 accuracy on MARS, and 96.00% Rank-1 and 100.0% Rank-5 accuracy on iLIDS-VID. We will release the code on GitHub after public release.

Takeaways, Limitations

•
Takeaways:
◦
We significantly improve the performance of video-based person re-identification by learning keypoint-based spatiotemporal representations.
◦
Effectively leverage holistic identity information and granular segment information through a combination of global and regional branches.
◦
Achieved state-of-the-art performance on MARS and iLIDS-VID benchmarks.
◦
Open code disclosure increases the reproducibility and usability of research.
•
Limitations:
◦
Since only performance on specific benchmark datasets is presented, generalization performance to other datasets requires further study.
◦
Performance may depend on the accuracy of keypoint extraction. Robustness studies on keypoint extraction errors are needed.
◦
There is a lack of analysis on real-time processing speed.
👍
Made with Slashpage