Code for "Ordinal Depth Supervision for 3D Human Pose Estimation", CVPR 2018
Switch branches/tags
Nothing to show
Clone or download
Latest commit 9b5b7f4 Jun 21, 2018
Permalink
Failed to load latest commit information.
full-eval Initial commit Jun 21, 2018
README.md Initial commit Jun 21, 2018
data.sh Initial commit Jun 21, 2018

README.md

Ordinal Depth Supervision for 3D Human Pose Estimation

Georgios Pavlakos, Xiaowei Zhou, Kostas Daniilidis

This is the code for the paper Ordinal Depth Supervision for 3D Human Pose Estimation. Please follow the links to read the paper and visit the corresponding project page.

We provide code to test our model on Human3.6M. Please follow the instructions below to setup and use our code. The typical procedure is 1) apply the ConvNet models using a torch script through command line and then 2) run a MATLAB script (from folder matlab) for visualization or evaluation. To run this code, make sure the following are installed:

We will soon update this repository with the training code.

1) Downloading models and data

We provide our models pretrained for Human3.6M. To get the models and the Human3.6M data to reproduce our results on this dataset, please run the following script (be careful, since the size is over 8GB)

bash data.sh

2) Evaluation on Human3.6M (sample)

We have provided a sample of Human3.6M images, following previous work. You can apply our model on this sample by running the command:

th main.lua h36m-sample

Then, to visualize the output, you can use the MATLAB script:

demoH36M.m

3) Evaluation on Human3.6M (full)

If you want to reproduce the results of our paper for Human3.6M, you need to download the full set of images we used for testing, by running the script data.sh as indicated above. These images are extracted from the videos of the original dataset, and correspond to the images used for testing by the most typical protocol. Please check the file:

data/h36m/annot/valid_images.txt

for the correspondence of images with the original videos.

Having downloaded the necessary images, to apply our model on the whole set of Human3.6M images, you can run:

th main.lua h36m

Then, to conclude the evaluation, you need to run the next MATLAB script:

evalH36M.m

This will print the output in a results.txt file.

Citing

If you find this code useful for your research, please consider citing the following paper:

@Inproceedings{pavlakos2018ordinal,
  Title          = {Ordinal Depth Supervision for 3{D} Human Pose Estimation},
  Author         = {Pavlakos, Georgios and Zhou, Xiaowei and Daniilidis, Kostas},
  Booktitle      = {Computer Vision and Pattern Recognition (CVPR)},
  Year           = {2018}
}

Acknowledgements

This code follows closely the released code for the Stacked Hourglass networks by Alejandro Newell. If you use this code, please consider citing the respective paper.