This is a PyTorch implementation of the following paper:

Learning a Delighting Prior for Facial Appearance Capture in the Wild., SIGGRAPH 2026.

Yuxuan Han, Xin Ming, Tianxiao Li, Zhuofan Shen, Qixuan Zhang, Lan Xu, and Feng Xu

Project Page | Video | Paper | Dataset

OpenDelight is a fully open-source, high-performance delighting prior tailored for facial appearance capture. It produces better results than the the best proprietary model, SwitchLight, with significantly fewer shadow-baking artifacts.

• Ava256-Scan Dataset. In this open-source project, we have processed the Ava256 dataset into the Light Stage scan format and made it publicly available. We name this dataset Ava256-Scan. Compared with the NeRSemble-Scan mentioned in the paper, this dataset features a larger scale, more comprehensive ethnic diversity, and achieves better performance for illumination removal networks trained on it. Therefore, we no longer plan to release the NeRSemble-Scan dataset.
• Lower Hardware Requirements. We have optimized the code at the engineering level. Now the full network training can be completed with only two 24GB GPUs such as the NVIDIA RTX 3090, and the 48GB GPUs required in the original paper are no longer necessary.

First, please follow the guidelines in doc/ENV.md to configure the environment for OpenDelight.

• (DONE) To reproduce the training pipeline of OpenDelight, refer to doc/TRAIN.md. It documents the entire workflow from rendering synthetic datasets to network training. Training OpenDelight to match the results reported in our paper is computationally cost-effective: it only takes 2 × NVIDIA RTX 3090 GPUs and converges within 3–4 days.

• (DONE) If you want to test our pre-trained model or leverage OpenDelight for facial appearance capture from a smartphone video (as we show in the paper), skip the training step directly and refer to doc/TEST.md.

• (TODO) Integrated with the WildCap system to build a fully automatic appearance capture pipeline.

If you have any questions or are interested in collaboration, please contact Yuxuan Han (hanyx22@mails.tsinghua.edu.cn).

Please include the following citations if it helps your research:

@inproceedings{han2026opendelight,
    author = {Han, Yuxuan and Ming, Xin and Li, Tianxiao and Shen, Zhuofan and Zhang, Qixuan and Xu, Lan and Xu, Feng},
    title = {Learning a Delighting Prior for Facial Appearance Capture in the Wild},
    booktitle = {SIGGRAPH},
    year={2026}
}