search

yerfor / GeneFacePlusPlus

GeneFace++: Generalized and Stable Real-Time 3D Talking Face Generation; Official Code
MIT License
1.56k stars 224 forks source link

training dataset issue 训练集的问题 #120

Open alphonz opened 8 months ago

alphonz commented 8 months ago

Thank you very much for this project. However, I have some questions regarding the training set:

  1. Is it necessary to be a single person in the dataset?
  2. Is it required to capture an undistorted 512*512 headshot?
  3. Is there a recommended duration for the footage?
tranthuhoai3786 commented 7 months ago

@alphonz there seems is no recommendation for duration. I used a video (39 second) and it worked fine

yueool commented 7 months ago

@alphonz

好像只有前10s有效，我看了代码

at: data_gen/utils/process_video/fit_3dmm_landmark.py line:417

for i in tqdm.trange(min(250, len(frames)), desc=f'rendering debug video to {debug_name}..'):

这是制作数据集中的代码：min(250, len(frames)) 因此我猜测只有前10秒有效，但是我并确定是这样，因为我的代码能力很差，更多是猜测

MohitPanpaliya commented 6 months ago

@alphonz there seems is no recommendation for duration. I used a video (39 second) and it worked fine

how much time it took to get the checkpoints file and train the model ?