Downloads

3D Hand Modeling

In many researches related to hand pose estimation a realistic 3D hand model is needed to measure the compatibility between input and hypothesized hand poses, and to visualize the prediction. In this project the hand model is modeled as a set of rigid quadratic surfaces, and has 27 degrees-of-freedom (DoFs) including 6 DoFs of global motion and 21 DoFs of local motion. The hand skeleton is modeled as a kinematic chain of 20 joints connected by bones in a tree structure with the root at the wrist. In this program you can view the model from different positions and perspectives, and manipulate each DoF separately. More importantly, to assist research in this area, we also provide functions to generate the depth image, silhouette image and edge images, which can be easily used for model-based matching between the input and the 3D hand model. This project is developed using C++/MFC/OpenGL/OpenCV. So make sure these libraries are installed before you run the project.

You can view the operations of the hand model in the video [here]. The full source code can be downloaded from my GitHub [Link].

Random Decision Forest for Per-pixel Hand Parsing

This project consists of the code for per-pixel classification with the Random Decision Forest. It includes a small sample dataset of around 100 annotated synthesized depth images, a program “SampleGenerator” to prepare the training data with 80% of these training images, and a program “RandomForest” for both Random Decision Forest training and testing on the remaining images. The code is written in C++/OpenCV. Please make sure you add the OpenCV library correctly before you run the code. For more details, please check the "Readme.txt" file and the program comments in the project. Please cite our below paper if you would use this project. Here is the donwload link [Link].

Hui Liang, Junsong Yuan and Daniel Thalmann, Parsing the Hand in Depth Images, in IEEE Trans. Multimedia, vol. 16, no. 5, Aug. 2014.

Here are some sample results for both per-pixel hand and body parsing with RDF. The body parsing dataset is publicly available from A. Hernandez-Vela's work in CVPR 2012 and can be obtained here [Link].

The code for some above projects can also be found in my Github account: https://github.com/shrekei