Spaces:
Running
与wav2lips的源代码和模型的区别
非常棒的工作。尤其是生产效率上非常的高，达到了实时的程度。
我想请问在工程方面与wav2lips的源代码和模型的区别是什么呢？ 从论文描述中，我发现图像重构的部分和wav2lips是一致的。
是否是再wav2lip的基础上进行了数据的训练产生新的模型，然后节后具体的人物模特进行了微调呢？
谢谢
Translated:
Great job. In particular, the production efficiency is very high, reaching a real-time level.
I would like to ask what is the difference between the source code and model of wav2lips in terms of engineering? From the description of the paper, I found that the image reconstruction part is consistent with wav2lips.
Is it based on wav2lip data training to generate a new model, and then fine-tuning the specific character models after the festival?Thanks
Hi @xingruispace ,
First of all, thank you for your interest in our demo!
There were questions about whether or not we succeeded in training Wav2Lip with the dataset from a single person.
Basically, we haven't started from the Wav2Lip code. We implemented our model from scratch and with PyTorch Lightning.
Also for the dataset, we record single-person video and training from scratch without any pre-training from LRS2, the dataset of Wav2Lip.
You can see the details of our model and how different it is from wav2lip paper at our model card: Is your demo made with Wav2Lip?.
Have a nice one.
非常感谢您详细的解答，以及对于模型的不同点的经验，祝您学习和工作顺利！
@xingruispace
고마워요 谢谢
Wish you good luck, too :)