Spaces:
Running
on
Zero
Update run_diffueraser.py
Hello Friends:
I wanted to propose a change to this line: ('--mask_dilation_iter', type=int, default=8, help='Adjust it to change the degree of mask expansion')
from default 8 to 2 (if possible, change it to 2)
because the quality of object or logo removal is extremely better.
I've tested it on other servers, even with objects on top of fast-moving videos, and the result is excellent.
Option 8 (DEFAULT) is fine when the objects to be removed are in a corner of the video, but when they are OVER the video.
Option 8 (DEFAULT) when the removal is complete, the video behind the object is deformed.
It deforms the scene behind the removed object.
If the video is moving fast, it's even worse, since the final result is deformed and looks bad.
Also, to save resources, you could change the default value of 960 in this line: ('--max_img_size', type=int, default=960, help='The maximum length of output width and height')
to 720, as this is sufficient to obtain a very good final result.
SUMMARY: Perhaps, although I don't know how DIFFUERASER works in extreme detail, more resources would be spent by improving object removal from 8 to 2, but resources would be gained by lowering the final video from 960 to 720, or even to 640.
THANK YOU VERY MUCH