search

mapooon / SelfBlendedImages

[CVPR 2022 Oral] Detecting Deepfakes with Self-Blended Images https://arxiv.org/abs/2204.08376
Other
196 stars 34 forks source link

Unreasonable choice of only youtube-real subset for preprocessing #45

Open molokanov50 opened 3 months ago

molokanov50 commented 3 months ago

Why do you limit your preprocessing with only youtube-real subset of FF++? Namely, you indicate -d Original parameter while running crop_dlib_ff.py and crop_retina_ff.py. And you throw NotImplementedError if a subset is not indicated. So how to preprocess the whole FF++ at a single run?