search

X-niper / UniTalker

Apache License 2.0
90 stars 7 forks source link

about trainning #14

Open ak01user opened 2 weeks ago

ak01user commented 2 weeks ago

hi,good jobs! Now I have a one-person dataset of flame with 5023*3 vertices each frames,about 250 frames,How can I train a new model using this dataset and to get more suitable results of this new person?Maybe it's a fine-tune work. thanks.

X-niper commented 2 weeks ago

hi,good jobs! Now I have a one-person dataset of flame with 5023*3 vertices each frames,about 250 frames,How can I train a new model using this dataset and to get more suitable results of this new person?Maybe it's a fine-tune work. thanks.

Sure, please finetune the released model and don't finetune it for long.

ak01user commented 2 weeks ago

Which part should I train specifically? I tried to modify the data loading part in train.py, load success! but it gave an error when training epoch 1. image

ak01user commented 2 weeks ago

it is successful when I set the dataset_name is "D1". I trained 2 epochs, Its result is good. But when I train, the fps is 25, and when I infer, I need to set it to 30 to align with the audio. Why is this?