ICFormer: Transformer with Inverse-Attention and Contrastive Learning for Polyp Segmentation

Datasets

1. You can refer to the work of MICCAI2020 https://github.com/DengPingFan/PraNet for related datasets.

2. Please follow the Section 4.1 of the paper to properly split your dataset.

3. After splitting the dataset, update the corresponding path in train.py and test.py accordingly.

Model Parameter Files Setup

1. Download the model parameter files:

   • The parameter will be released as soon.

2. Place the downloaded mit_b4.pth and The_best_Epoch.pth files in the appropriate paths:

   • Place mit_b4.pth in the lib/backbone/ (recommended path).
   • Place The_best_Epoch.pth in the experiment/exp_icformer_1/ (recommended path).

3. (Optional) You can update the file paths in the code for these parameters:

   • In /lib/network/network_demo2.py, set the path for mit_b4.pth.
   • In test.py, set the path for The_best_Epoch.pth.