Closed Mohamed209 closed 4 years ago
What is the input source?
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
Try to crop the target video, it is too large probably.
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
Yes.
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