The model weights used in our paper are available at Hugging Face Model.
| Train dataset | Val dataset/Subset/Mask | Checkpoint | Config | Test set/Subset/Mask | MAD↓ | Grad↓ | dtSSD↓ |
|---|---|---|---|---|---|---|---|
| I-HIM50K | HIM2K+M-HIM2K/comp/r50_fpn_3x | chuonghm/maggie-image-him50k-cvpr24 | maggie_image.yaml | HIM2K+M-HIM2K/natural/all | 27.17 | 9.94 | - |
| V-HIM2K5 | V-HIM60/comp_medium/xmem | chuonghm/maggie-video-vim2k5-cvpr24 | maggie_video.yaml | V-HIM60/comp_hard/xmem | 21.23 | 7.08 | 29.90 |
We provide baseline's weights in Google Drive. Please check BASELINES for training and evaluation. For quantitative results, please check the paper
Image baselines include:
Video matting include: