We curated a subset of test subjects from the CelebA-HQ and FFHQ datasets for the quantitative comparisons against baseline methods in our paper. Specifically, we selected:
For each test subject, we generated a corresponding segmentation mask, which is designed to keep the eye and mouth areas visible when needed.
The test subject lists… See the full description on the dataset page:
https://huggingface.co/datasets/hkung/nullface-test-set.
