Use the package manager pip to install pipenv, a virtual environment tool

pip install pipenv

then install the requirements :

cd ............/django_project/django_project/

pip install -r requirements.txt

After that, we need to install CUDA toolkit 10

finally, a depth sensor is required ! I'm using Orbbec Astra Mini with openNI sdk. You need to install the suitable driver.

The libraries are integrated within a django project to use it as a backend to use it for an interactive projection solution, but you can use it separetly using the following commands :

cd ............/3D_hand_pose_system/
pipenv shell
cd django_project/
python handpose.py

I'm providing the resulted pre-trained weights Please download them to /django_project/Hand_pose_estimation_part/model

The model is based on the great work of Kuo Du and others : Paper and Code. A great thanks for sharing their work with public.

For this project, the model is trained using NYU dataset (this version contains only the needed files, 4GB instead of 92GB) , please follow the instructions which I have provided in the training notebook.

At first, we used a pre-trained RGB based hand detector but as the light emitted by the data show affected badly its performance, we managed to train another one using the depth sensor, details are in the associated notebook.

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

CrossInfoNet: Multi-Task Information Sharing Based Hand Pose Estimation Kuo Du1 Xiangbo Lin2* Yi Sun2 Xiaohong Ma2 Dalian University of Technology, China

MIT