we use Talking-Head-1K-Hour as the example.
/home/xxx/TH1KH_512/video_raw/*.mp4
data_gen/utils/process_video/resample_video_to_25fps_resize_to_512.py
/home/xxx/TH1KH_512/video/*.mp4
data_gen/utils/process_video/extract_segment_imgs.py
/home/xxx/TH1KH_512/{gt_imgs, head_imgs, inpaint_torso_imgs, com_imgs}/*
data_gen/utils/process_video/extract_lm2d.py
/home/xxx/TH1KH_512/lms_2d/*_lms_2d.npy
data_gen/utils/process_video/fit_3dmm_landmark.py
/home/xxx/TH1KH_512/coeff_fit_mp/*_coeff_fit_mp.npy
data_gen/utils/process_audio/extract_mel_f0.py
/home/xxx/TH1KH_512/audio/*.wav
and mel_f0 in /home/xxx/TH1KH_512/mel_f0/*_mel_f0.npy
data_gen/utils/process_audio/extract_hubert.py
/home/xxx/TH1KH_512/hubert/*_hubert.npy
data_gen/runs/binarizer_th1kh.py
data/binary/th1kh