search

AliaksandrSiarohin / motion-cosegmentation

Reference code for "Motion-supervised Co-Part Segmentation" paper
Other
653 stars 142 forks source link

output video from partswap.py is very blurry #24

Closed Mohamed209 closed 4 years ago

Mohamed209 commented 4 years ago

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

Screenshot_2020-08-27_02-18-26

In other words , how can I get higher resolution results

AliaksandrSiarohin commented 4 years ago

What is the input source?

Mohamed209 commented 4 years ago

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 commented 4 years ago

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

Mohamed209 commented 4 years ago

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 commented 4 years ago

Yes.