Cross-Modal Deep Variational Hand Pose Estimation

Project page

This repository provides a code base to evaluate the trained models of the paper Cross-Modal Deep Variational Hand Pose Estimation and reproduce the numbers of Table 2. It is a modified version of the code found here by Christian Zimmermann, adapted to run our model.

Recommended system

Recommended system (tested):

Ubuntu 16.04.2 (xenial)
Python 3.5.2

Python packages used by the example provided and their recommended version:

tensorflow==1.3.0
numpy==1.14.5
scipy==1.1.0
matplotlib==1.5.3
pytorch==0.3.1
opencv-python==3.4.1.15

Preprocessing for training and evaluation

In order to use the training and evaluation scripts you need download and preprocess the datasets.

Rendered Hand Pose Dataset (RHD)

Download the dataset RHD dataset v. 1.1
Extract it.
In the file 'create_binary_db.py', set the variable 'path_to_db' to the path of the extracted dataset.
Run

python create_binary_db.py

This will create a binary file in ./data/bin according to how 'set' was configured. Keep it at 'evaluation'.

Stereo Tracking Benchmark Dataset (STB)

To run the dataset on STB, it is neecessary to get the dataset presented in Zhang et al., ‘3d Hand Pose Tracking and Estimation Using Stereo Matching’, 2016
The link to the dataset can be found here
Unzip the dataset
In 'create_db.m', set 'PATH_TO_DATASET' to the path of the extracted dataset
Run

cd ./data/stb/
matlab -nodesktop -nosplash -r "create_db"

This will create the binary file ./data/stb/stb_evaluation.bin

Trained models

The trained models can be found here.
After downloading the models, unzip it into the main project directory.

Evaluation

To reproduce the numbers of table 2, set the desired experimental settings in 'evaluate_model.py'. These correspond to the following parameters in the paper:

Paper	Code
H	hand_side_invariance
S	scale_invariance

python evaluate_model.py

This will create a folder named 'eval_results' in the respective model folder in which the PCK curve and some predictions are saved.

License and Citation

This project is licensed under the terms of the GPL v2 license. By using the software, you are agreeing to the terms of the license agreement.

If you use this code in your research, please cite us as follows:

@inproceedings{spurr2018cvpr,
    author = {Spurr, Adrian and Song, Jie and Park, Seonwook and Hilliges, Otmar},
    title = {Cross-modal Deep Variational Hand Pose Estimation},
    booktitle = {CVPR},
    year = {2018},
    location = {Salt Lake City, USA},
}

liubl1217/vae-hands-3d