moberweger / deep-prior

Fast and accurate 3D hand pose estimation from single depth images

This work is superseded by DeepPrior++

DeepPrior - Accurate and Fast 3D Hand Pose Estimation

Author: Markus Oberweger [e-mail]

Requirements:

• OS
  • Ubuntu 14.04
  • CUDA 7
• via Ubuntu package manager:
  • python2.7
  • python-matplotlib
  • python-scipy
  • python-pil
  • python-numpy
  • python-vtk6
  • python-pip
  • python-vtk6
• via pip install:
  • scikit-learn
  • progressbar
  • psutil
  • theano (0.8)
• Camera driver
  • OpenNI for Kinect
  • DepthSense SDK for Creative Senz3D.

For a description of our method see:

M. Oberweger, P. Wohlhart, and V. Lepetit. Hands Deep in Deep Learning for Hand Pose Estimation. In Computer Vision Winter Workshop, 2015.

Setup:

• Put dataset files into ./data (e.g. ICVL dataset, or NYU dataset )
• Goto ./src and see the main file test_realtimepipeline.py how to handle the API
• Camera interface for the Creative Senz3D is included in ./src/util. Build them with cmake . && make.

Pretrained models:

Download pretrained models for ICVL and NYU dataset.

Datasets:

The ICVL dataset is trained for a time-of-flight camera, and the NYU dataset for a structured light camera. The annotations are different. See the papers for it.

D. Tang, H. J. Chang, A. Tejani, and T.-K. Kim. Latent Regression Forest: Structured Estimation of 3D Articulated Hand Posture. In Conference on Computer Vision and Pattern Recognition, 2014.

J. Tompson, M. Stein, Y. LeCun, and K. Perlin. Real-Time Continuous Pose Recovery of Human Hands Using Convolutional Networks. ACM Transactions on Graphics, 33, 2014.