Pytorch version of ‘How far are we from solving the 2D \& 3D Face Alignment problem? (and a dataset of 230,000 3D facial landmarks)’
For official torch7 version please refer to face-alignment-training
This is a reinplement of training code for 2D-FAN and 3D-FAN decribed in “How far” paper. Please visit author’s webpage or arxiv for technical details.
Thanks for bearpaw’s excellent work on human pose estimation pytorch-pose . And in this project, I reused a branch of helper function from pytorch-pose.
Pretrained models are available soon.
- Install the latest PyTorch, version 0.2.1 is fully supported and there is no further test on older version.
- scipy
- torchvision
- progress(optional) for better visualization.
- Clone the github repository and install all the dependencies mentiones above.
git clone https://github.com/hzh8311/pyhowfar
cd pyhowfar
- Download the 300W-LP dataset from the authors webpage. In order to train on your own data the dataloader.lua file needs to be adapted.
- Download the 300W-LP annotations converted to t7 format by paper author from here, extract it and move the “`landmarks“` folder to the root of the 300W-LP dataset.
In order to run the demo please download the required models available bellow and the associated data.
python main.py
In order to see all the available options please run:
python main.py --help
- Pythoner friendly and there is no need for `.t7` format annotations
- Add 300-W-LP test set for validation.
- Followed the excatly same training procedure described in the paper (except binary network part).
- Add model evaluation in terms of **Mean error**, **[email protected]**
- TODO: add evaluation on test sets (300W, 300VW, AFLW2000-3D etc.).
@inproceedings{bulat2017far,
title={How far are we from solving the 2D \& 3D Face Alignment problem? (and a dataset of 230,000 3D facial landmarks)},
author={Bulat, Adrian and Tzimiropoulos, Georgios},
booktitle={International Conference on Computer Vision},
year={2017}
}