LAMP

This repo is the official implementation for: "LAMP: Leveraging Language Prompts for Multi-person Pose Estimation" @IROS2023

Method

Requirements

• python 3.9
• pytorch 1.12.1
• torchvision 0.13.1
• clip 1.0

Datasets

• OCHuman
• CrowdPose

Training

• To train on OCHuman from ImageNet pretrained models with multiple gpus, you can run:

python3 tools/train.py --cfg experiments/ochuman.yaml --gpus 0,1,2,3

where --cfg indicates the configure file and --gpus implys the numers of gpus. You can replace the configure file for other datasets.

Testing

• To test the model which has been trained on the OCHuman dataset, you can run the testing script as following:

python tools/valid.py --cfg experiments/ochuman.yaml --gpus 0,1,2,3 TEST.MODEL_FILE MODEL_PATH/model_best.pth.tar

Replace MODEL_PATH with your local path of the trained model.

Citation

Please cite the following paper if you find this repository useful in your research.

@misc{hu2023lamp,
      title={LAMP: Leveraging Language Prompts for Multi-person Pose Estimation}, 
      author={Shengnan Hu and Ce Zheng and Zixiang Zhou and Chen Chen and Gita Sukthankar},
      year={2023},
      eprint={2307.11934},
      archivePrefix={IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
      primaryClass={cs.CV}
}

Reference

The code is mainly encouraged by

• CID
• MMPose
• CLIP