can I improve the quality of the resulting video? And how to do this?

[Vova-B]

I want to receive videos in higher quality and resolution, what steps do I need to take to get the desired result?

[DBDXSS]

Change the train_width, train_height, and fps in configs/train/stage2.yaml to your desired augments. Then finetune the model.

If you want to reference only, change the augments in configs/inference/default.yaml

[Vova-B]

@DBDXSS For finetuning, do I need to run only train stage 2? If I start training with train stage 1, then it does not fine-tune the model but trains from 0. Tell me how to launch fine-tuning

[DBDXSS]

@Vova-B I am not a developer of this project, so I won't be able to give you a definite conclusion. The results of the two methods cannot indicate which one is better. You can also use weights for initialization of the training stage 1, in fact. Since the facial features of your fine-tuned data and the author's training data may be different, I think it will be better to fine-tune from stage1. However, this training process will take a lot of time. You can also fine-tune directly from stage2. If the effect is not good enough, then try starting from stage1 again.