Hand detection in PyTorch

• Installation
• Training
• Demo
• References

1. Install PyTorch-0.4.0 according to your environment.

2. Clone this repository. We will call the cloned directory as $HandBoxes_ROOT.

git clone https://github.com/zllrunning/hand-detection.PyTorch.git

3. Compile the nms:

./make.sh

Note: We currently only support PyTorch-0.4.0 and Python 3+.

1. Prepare training data:

	-- download EgoHands dataset
	-- generate bounding boxes and visualize them to ensure correctness
	-- convert bbox file to VOC format

cd $HandBoxes_ROOT/
sh prepare_data.sh

2. Train the model using EgoHands dataset:

python3 train.py

If you do not wish to train the model, you can download our pre-trained model and save it in $HandBoxes_ROOT/weights.

1. Evaluate the trained model using:

# evaluate using GPU
python test.py --video data/video/hand.avi
# evaluate using cpu
python test.py --image data/video/CARDS_OFFICE_H_T_frame_1085.jpg --cpu

This project is based on FaceBoxes.PyTorch

• handtracking
• od-annotation