recommended preprocessing for video model about modnet HOT 4 CLOSED

there is still some difference from https://github.com/ZHKKKe/MODNet/blob/master/demo/image_matting/colab/inference.py where the logic is applied only when if max(im_h, im_w) < ref_size or min(im_h, im_w) > ref_size. is that more appropriate for the photographic model but not for the video model?

from modnet.

Hi, thanks for your attention.

Q1: for the pretrained webcam/custom video model, is it recommended to always resize to 672x512 ?
Nope. I think the original video ratio should be used. Sorry that We forgot this point before.
We will work on this and update the demo in these two days.

from modnet.

@eyaler The bug is fixed. Thanks,

from modnet.

Nope. I found that the effect of the conditional statement if max(im_h, im_w) < ref_size or min(im_h, im_w) > ref_size on the result is negligible. Therefore, I did not add it in the video demo.

from modnet.