ddshan / hand_detector.d2

Hand detection models trained on 100DOH (100 Days of Hands) dataset.

Hand Detector

The hand detectors are trained on (1) 100K and (2) 100K+ego images from 100DOH dataset.

Performance

Name	Data	Box AP	Model
Faster-RCNN X101-FPN	100K	90.32%	Google Drive
Faster-RCNN X101-FPN	100K+ego	90.46%	Google Drive

Environment

• Set up detectron2 environment as in install.md

Train

CUDA_VISIBLE_DEVICES=4,5,6,7 python trainval_net.py --num-gpus 4 --config-file faster_rcnn_X_101_32x8d_FPN_3x_100DOH.yaml

Evaluation

CUDA_VISIBLE_DEVICES=4,5,6,7 python trainval_net.py --num-gpus 4 --config-file faster_rcnn_X_101_32x8d_FPN_3x_100DOH.yaml --eval-only MODEL.WEIGHTS path/to/model.pth

Demo

CUDA_VISIBLE_DEVICES=1 python demo.py

Citation

If this work is helpful in your research, please cite:

@INPROCEEDINGS{Shan20, 
    author = {Shan, Dandan and Geng, Jiaqi and Shu, Michelle  and Fouhey, David},
    title = {Understanding Human Hands in Contact at Internet Scale},
    booktitle = CVPR, 
    year = {2020} 
}

When you use the model trained on our ego data, make sure to also cite the original datasets (Epic-Kitchens, EGTEA and CharadesEgo) that we collect from and agree to the original conditions for using that data.