Yuval Nirkin
Yuval Nirkin
Can you try with CUDA 10.1 or CUDA 11?
Target images are not yet supported in v2. For now you can use [face_swap_images2images.py](https://github.com/YuvalNirkin/fsgan/blob/v1/inference/face_swap_images2images.py) from v1.
Yes, please refer to [this guide](https://github.com/YuvalNirkin/fsgan/wiki/Paper-Models-Inference).
[swap.py](https://github.com/YuvalNirkin/fsgan/blob/master/inference/swap.py) supports single images as well.
It was a quick and dirty experiment and not really tested thoroughly . It might be that faces were not detected in some of the frames.
It might be that no faces were detected for one of the videos. Try with different videos (maybe easier ones) and see if the error still occurs.
Maybe you don't have ffmpeg installed on your system. Follow the installation guide on how to install it.
Overall changing the shape of the target face is not very well defined. This face swapping methods follows the face landmarks of the target face and therefore does not preserve...
Thank you for the demo. It didn't work for me with large videos, it might need a little more refinement before I can add a link from the repository.
Just call the script with lower values for all batch* arguments, most importantly the "det_batch_size" argument.