yutao1008 / margin_calibration

A new designed learning objective for deep image segmentation.

Pytorch implementation of COPLE-Net with the proposed margin calibration method.

Running dependencies: Python 3.7 (Anaconda) Pytorch 1.1 tqdm

Before running the code, please edit the values of root in datasets/covid19_lesion.py and datasets/robotic_instrument.py, which are the image/data directories.

To train the segmentation model, simply run the command with the arguments like follows:

python -W ignore train.py
--dataset robotic_instrument
--task parts
--margin_loss
--date 0907
--batch_size 6
--max_epoch 50
--adamw
--lr 1e-4
--exp robotic_instrument_parts_mg

To evaluate the model, just run:

python -W ignore eval.py
--dataset robotic_instrument
--task parts
--dump_imgs
--method mg
--snapshot <MODEL_CHECKPOINT_PATH>

If you feel this work useful to your research, please kindly cite the paper as follows:

@article{yu2022distribution,
        title={Distribution-Aware Margin Calibration for Semantic Segmentation in Images},
        author={Yu, Litao and Li, Zhibin and Xu, Min and Gao, Yongsheng and Luo, Jiebo and Zhang, Jian},
        journal={International Journal of Computer Vision},
        volume={130},
        number={1},
        pages={95--110},
        year={2022},
        publisher={Springer}
        }