A Benchmark for face hallucination in low-light scenarios
Face hallucination in low-light environments is an extremely challenging task due to the significant loss of facial structure and facial texture information. We construct a low-light face hallucination dataset DARKFFHQ-4000 based on the face dataset FFHQ as a benchmark for low-light face hallucination task. This dataset will provide researchers with a standardized and uniform benchmark for training and evaluation.
We select 4000 face images from the FFHQ dataset and downsample the HR face images using the Bicubic degradation model with scale factors of 4 and 8, resulting in 4000 face images with 64
The dataset can be downloaded from Google Drive, which contains HR, LR (scale factors are 4 and 8), and low-light LR face images.
If you find this dataset helpful in your research or work, please cite the following paper.
@article{wang2023learning,
title={Learning to Hallucinate Face in the Dark},
author={Wang, Yuanzhi and Lu, Tao and Yao, Yuan and Zhang, Yanduo and Xiong, Zixiang},
journal={IEEE Transactions on Multimedia},
year={2023},
publisher={IEEE}
}