Can preprocess.py not be used?

[cncbec]

In the new version of training, can this step be skipped? python preprocess.py

[lililuya]

Hi， You should use python preprocess.py to get frames and audio files which is necessary input-data for the training

[cncbec]

@lililuya thank you for your reply. But what about the command parameters including add_argument generated by preprocesed that are not added to the train_syncnet_sam.py and hq_wav2lip_sam_train.py files? Could it be that the train.txt of train_dataset = Dataset('filelists/train.txt') is the image and audio path generated by preprocesed?

[lililuya]

Hi, as the filelists you mentioned, you have to code a script to generate your own dataset filelists（just simple absoluted files' path）, you can refer the lrs2 dataset's filelists as well. Btw, you can change the get_image_list() function in you own way to get the data without filelists. I think you should refer the original project to understand this part.

[cncbec]

Thank you,I will try agian.