---
thumbnail: "https://github.com/VlSomers/keypoint_promptable_reidentification/blob/main/assets/keypoint_promptable_reid.jpg"
tags:
- reid
---
# KPR: Keypoint Promptable Re-Identification
[[Paper](https://arxiv.org/abs/2407.18112)] [[GitHub](https://github.com/VlSomers/keypoint_promptable_reidentification)] [[Demo](torchreid/demo.py)] [[Video](https://www.youtube.com/watch?v=IatrRfXrdnE)] [[Poster](assets/ECCV24_KPR_poster.pdf)] [[BPBreID](https://github.com/VlSomers/bpbreid)] [[TrackLab](https://github.com/TrackingLaboratory/tracklab)]

[![arXiv](https://img.shields.io/badge/arXiv-2407.18112-<COLOR>.svg)](https://arxiv.org/abs/2407.18112) [![Hippocratic License HL3-LAW-MEDIA-MIL-SOC-SV](https://img.shields.io/static/v1?label=Hippocratic%20License&message=HL3-LAW-MEDIA-MIL-SOC-SV&labelColor=5e2751&color=bc8c3d)](https://firstdonoharm.dev/version/3/0/law-media-mil-soc-sv.html)

🔥 *SOTA ReID model, with optional keypoint prompts, that is robust to multi-person occlusions* 🔥

✅ *Plug-and-play with any pose estimator or through manual keypoint selection*

🚀 *Release of the **Occluded-PoseTrack-ReID** dataset with manual keypoint annotations + keypoints annotations for four existing ReID benchmarks* 

KPR can be easily integrated into any codebase to perform person retrieval, person search, multi-person (pose) tracking, multi-view multi-skeleton matching for 3D pose, ...

SOTA performance with the [SOLIDER](https://github.com/tinyvision/SOLIDER) human-centric foundation model as backbone.


>**[Keypoint Promptable Re-Identification, ECCV24](https://arxiv.org/abs/2407.18112)**
>
>Vladimir Somers, Alexandre Alahi, Christophe De Vleeschouwer
>
>[*arxiv 2407.18112*](https://arxiv.org/abs/2407.18112)
>