Free-view Face Relighting using a Hybrid Parametric Neural Model on a SMALL-OLAT Dataset

Our results. This repository contains our implementation for the paper: Free-view Face Relighting using a Hybrid Parametric Neural Model on a SMALL-OLAT Dataset. We present a hybrid parametric-neural relighting (PN-Relighting) technique. Taking a single portrait image as input (a), we generate the surface geometry and albedo (b) and a free-view 3D face relightable (c) under different illumination.

Installation

Tested on Pytorch 1.11.0

Install environment:

conda create -n PNRelight python=3.8
conda activate PNRelight
conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=11.3 -c pytorch
pip install tqdm scikit-image opencv-python configargparse lpips imageio-ffmpeg Ninja tensorboard

Run

This directory contains the code for running relighting module in our framework.

The sub-folders are:

  • networks: Contains scripts for our networks mentioned in the paper
  • utils: Contains utils in our implementation
  • eval: Contains scripts for running our code and getting the outputs
  • test_data:
    • images: Contains test images at 512px
    • masks: Contains masks at 512px, corresponding to the images
    • env: Contains environment map, corresponding to each image for relighting
  • checkpoints: Folder to store checkpoints

In order to run our code, you have to download the necessary model checkpoints. We provide instructions to download checkpoints:

Finally, execute eval/normal_test.py to get surface geometry, eval/albedo_test.py to get albedo map and eval/relit_test.py to get relighting result. We provide examples below for you to try. We also provide eval/gen_demo.py that goes over all combinations and stores the outputs under test_data/relight.

You can use the following specifications:

  • python eval/gen_demo.py --folder=test_data

Citation

If you find our code or paper helps, please consider citing:

@article{Wang2023,
author = {Wang, Youjia and He, Kai and Zhou, Taotao and Yao, Kaixin and Li, Nianyi and Xu, Lan and Yu, Jingyi},
doi = {10.1007/s11263-022-01730-5},
issn = {1573-1405},
journal = {International Journal of Computer Vision},
number = {4},
pages = {1002--1021},
title = {{Free-view Face Relighting Using a Hybrid Parametric Neural Model on a SMALL-OLAT Dataset}},
url = {https://doi.org/10.1007/s11263-022-01730-5},
volume = {131},
year = {2023}
}