AIM
AIM copied to clipboard
No predictions for smaller objects
Hi, Thanks for the interesting work!
There are no predictions for samples that have smaller objects. Any suggestions that may help? Tested with test_samples.sh. Trying your pretrained model transfer learnt with DUTS. Example image below:
Hi there,
Thanks for showing interest in our work. As the released model is trained on the typical matting composite dataset with matting foregrounds occupying most of the part, you may get better results on small objects images if you re-train the code on customed small objects dataset. The training code has all been released for your convenience. Thanks.
Hi guys, how did you predict the image? There is no code sample.