search

Rudrabha / Wav2Lip

This repository contains the codes of "A Lip Sync Expert Is All You Need for Speech to Lip Generation In the Wild", published at ACM Multimedia 2020. For HD commercial model, please try out Sync Labs
https://synclabs.so
10.28k stars 2.21k forks source link

eval time #338

Open Carlyx opened 2 years ago

Carlyx commented 2 years ago

Hi, Your work is amazing!

Q1: I train the Wav2Lip model with LRS2. But during the training process, "eval_model"(https://github.com/Rudrabha/Wav2Lip/blob/b9759a3467cb1b7519f1a3b91f5a84cb4bc1ae4a/wav2lip_train.py#L251) takes about 30 minutes (on one V100), is this normal?

Q2: In LRS2, the train.txt has 45839 lines, and the batch_size of train_data_loader is 64. So len(train_data_loader) = 45839/64 =717. But in the loop of https://github.com/Rudrabha/Wav2Lip/blob/b9759a3467cb1b7519f1a3b91f5a84cb4bc1ae4a/wav2lip_train.py#L210, I got an output of over 717. L1: 0.026415626978400907, Sync Loss: 0.0: : 2865it [2:04:31, 3.36it/s] I'm not particularly sure why this is happening, any advice for me please?

sahilg06 commented 2 years ago

Hi @Carlyx, Question 2) If you see carefully, there are 2 batch_sizes in the hparams.py, So the batch_size you are using is 16, not 64, that's why its showing 2865.

Question 1) You may try changing this eval_steps variable here