This is the official repo for ECCV2024 paper "You Only Learn One Query: Learning Unified Human Query for Single-Stage Multi-Person Multi-Task Human-Centric Perception".

The repo contains COCO-UniHuman annotations and HumanQueryNet proposed in this paper.

2024/08/09: code and model weight of HumanQueryNet released!

2024/07/09: COCO_UniHuman dataset released!

Please refer to the introduction of dataset COCO_UniHuman.

conda create -n HQN python==3.9

conda activate HQN

pip install -r requirements.txt

1. Download COCO'17 images and COCO-UniHuman v1 annotations, add data_prefix and anno_prefix to the data config file configs/coco_unihuman_v1.py

2. Download the converted SMPL models from download link and put all files in HumanQueryNet/models/smpl/models:

HumanQueryNet/models/smpl/models/
├── gmm_08.pkl
├── SMPL_FEMALE.pth
├── SMPL_MALE.pth
└── SMPL_NEUTRAL.pth

3. Then modify train.sh to train the model (Please refer to mmdet-2.5.3 training scripts).

Our r50 model can be downloaded here.

Please refer to test.sh to test the model on all HCP tasks.

Codes and data are freely available for free non-commercial use, and may be redistributed under these conditions. For commercial queries, please contact Mr. Sheng Jin (jinsheng13[at]foxmail[dot]com). We will send the detail agreement to you.

if you find our paper and code useful in your research, please consider giving a star and citation:

@inproceedings{jin2023you,
  title={You Only Learn One Query: Learning Unified Human Query for Single-Stage Multi-Person Multi-Task Human-Centric Perception},
  author={Jin, Sheng and Li, Shuhuai and Li, Tong and Liu, Wentao and Qian, Chen and Luo, Ping},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2024},
  month={September}
}