SRN: Stacked Regression Network for Real-time 3D Hand Pose Estimation [paper] [supplementary]

Realtime demo

normal hand

small hand

^{* Demos above are realtime results from Kinect V2 using models trained on Hands17 dataset (Intel Realsense SR300).}

Usage

Prepare pre-trained models

Download models from here.

Put checkpoint/ in realtime/

Prepare test data

We provide some data collected by Kinect V2 in realtime/data/kinect2 for testing.

(Currently only Kinect V2 and realsense R300 are supported. If you want to use the depth map obtained by other sensors, you can configure the corresponding sensor parameters in run.py)

Testing

cd realtime
python run.py --data_dir ./data/your_folder_name --test_dataset kinect2 or realsense  

Comparison with state-of-the-art methods

NYU

ICVL

MSRA

HANDS17