output video from partswap.py is very blurry

[Mohamed209]

check attached image , output of face swap is very blurry (image taken from video)
the question is does this relate to networks (segmentation network or generator) operations , or relates to frame post or pre-processing after model predictions

In other words , how can I get higher resolution results

[AliaksandrSiarohin]

What is the input source?

[Mohamed209]

What is the input source?

here is original target video : https://drive.google.com/file/d/1QSEECWaH3J-KrYV1CG13at7t92tIIkE_/view?usp=sharing
and input source : https://drive.google.com/file/d/1DWAvcKoVCGOaM1Ywoa6ulsCTQu8QqMQD/view?usp=sharing

I have also checked all post-processing steps , i think resolution loss happens in model predictions stage !

and here is my command !python part_swap.py --source_image 'image path' --target_video 'target.mp4' --checkpoint weights/vox-cpk.pth.tar --config config/vox-256-sem-10segments.yaml --supervised --first_order_motion_model --swap_index 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15

[AliaksandrSiarohin]

Try to crop the target video, it is too large probably.

[Mohamed209]

Try to crop the target video, it is too large probably.

shall this script crop desired speaker face in all video frames to generate smaller version of the original video !

https://github.com/AliaksandrSiarohin/first-order-model/blob/master/crop-video.py

[AliaksandrSiarohin]

Yes.