我感觉这行代码应该是错了

[nghiakvnvsd]

我感觉这行代码应该是错了

[DYJNG]

+1

[LiuJiaqi0505]

https://github.com/nghiakvnvsd/wav2lip384/blob/13d9532e4b220a9ea930412b8a701ef2a2963d2f/parallel_wav2lip_margin.py#L195C18-L195C18

应该改成 vidname_mel = self.all_videos[idx] 否则 训练的音频视频不是同一个文件.

right. i have found the same problem.

[nghiakvnvsd]

Hi everyone,

Sorry for replying to you late.
This repo is modified directly from the original wav2lip repo, and it requires a large amount of high-resolution data to train. So it's obvious that you won't train well. But currently, I have a strategy to train the target person and it only requires about 40 hours of data. Please contact me if your company needs a full-time data scientist to deploy this model and wants to research better models. This code just committed to finding a new job. I will remove the code when I get a job. If your company wants to offer a full-time job for the lipsync project, please get in touch with me. I will show my results and when I get the offer, I will deploy my full strategy for your company. I have other plans in the future to deploy other models such as GeneFace and MemFace but I lack the resources to research further.

[zhangbo2008]

我改了一版 https://github.com/zhangbo2008/wav2lip384_my2
i have a modified version.