This project is the tensorflow implementation of VNect: Real-time 3D Human Pose Estimation with a Single RGB Camera, SIGGRAPH 2017.
This is not an official implementation. Please contact paper author for related model.
- Ubuntu 16.04
- Python 2.7
- Tensorflow 1.3.0
- OpenCV 3.3.0
- OpenGL (optional)
- 1.Download model, put them in folder
models/weights - 2.Edit demo settings in shell script,
--device--demo_type--model_file--test_img--plot_2d--plot_3d - 3.If you have OpenGL, you can run
run_demo_tf_gl.shfor faster rendering of 3d joints. Otherwise, runrun_demo_tf.sh
- Some bugs in detected 3D joint locations.
- Training part of model.