The offical repo for Trimap-guided feature mining and fusion network for natural image matting.
pip install -r requirement_new.txtpython3 -m torch.distributed.launch --nproc_per_node=4 --master_port=$PORT \
tools/train.py configs/mattors/gradloss/tmflaploss020.py --launcher pytorch --work-dir $WORKDIR --ckpt-least 190000 --eval-least 500000 --eval-interval 2000 --ckpt-interval 2000 --total-iters 200000 --per-gpu 16| Model | Training set | Test set | TTA | SAD | MSE | GRAD | CONN | Download |
|---|---|---|---|---|---|---|---|---|
| TMF_comp1k | Composition-1K train | Composition-1K test | No | 23.0 | 4.0 | 7.5 | 18.7 | BaiduYun(Access Code:gjjr) Google Drive |
| TMF_comp1k | Composition-1K train | Composition-1K test | Yes | 22.1 | 3.6 | 6.7 | 17.6 | as above |
| TMF_ciom | CIOM train | CIOM test | No | 20.2 | 1.8 | 4.8 | 13.6 | BaiduYun(Access Code:zcww) Google Drive |
| TMF_ciom | CIOM train | Composition-1K test | No | 21.6 | 4.0 | 7.6 | 17.1 | as above |
| TMF_ciom | CIOM train | Composition-1K test | Yes | 20.8 | 3.8 | 6.7 | 16.0 | as above |
./tools/dist_test.sh configs/mattors/gradloss/tmflaploss020.py comp1k.pth 2
###or with TTA
./tools/dist_test.sh configs/mattors/gradloss/tmflaploss020tta8.py comp1k.pth 2If you find TMFNet useful in your research, please consider citing:
@article{jiang2023trimap,
title={Trimap-guided feature mining and fusion network for natural image matting},
author={Jiang, Weihao and Yu, Dongdong and Xie, Zhaozhi and Li, Yaoyi and Yuan, Zehuan and Lu, Hongtao},
journal={Computer Vision and Image Understanding},
volume={230},
pages={103645},
year={2023},
publisher={Elsevier}
}