TMElyralab / MusePose

MusePose: a Pose-Driven Image-to-Video Framework for Virtual Human Generation
Other
2.15k stars 153 forks source link

some questions about training #39

Open jeffreyking23 opened 4 months ago

jeffreyking23 commented 4 months ago

positional_encoding_max_len is 128 in inference config, but in the inference slice_number default number is 48, i am confusing about that , is slice_number equal to 128 in the training?

TZYSJTU commented 4 months ago

positional_encoding_max_len is 128 in inference config, but in the inference slice_number default number is 48, i am confusing about that , is slice_number equal to 128 in the training?

128 is the max length, which means your can train a motion module to smooth video segments with less than 128 frames. During training, we use n_sample_frames = 48. If positional_encoding_max_len < n_sample_frames, the code will throw a error.

Jaxkr commented 3 months ago

Where are you getting training code?