WeakPolyp-SAM: Segment Anything Model-driven Weakly-supervised Polyp Segmentation

Authors:

Yiming Zhao, Tao Zhou, Yunqi Gu, Yi Zhou, Yizhe Zhang, Ye Wu, and Huazhu Fu.

• This repository provides code for "WeakPolyp-SAM: Segment Anything Model-driven Weakly-supervised Polyp Segmentation"

Figure 1: Overview of the proposed framework.

Figure 2: Visualization results of different methods on the polyp segmentation.

1. Installation

   To set up the environment and install dependencies, run:

   pip install -r requirements.txt

2. Data:

   • downloading Polyp dataset, which can be found from GoogleDrive, run:

   unrar x PolypData.rar

   • downloading SAM weights and move it into ./checkpoints/sam_vit_b_01ec64.pth, which can be found from GitHub.

3. Preprocessing:

   • Download PolypData and put them into ./data folder with the following structure:

     |-- data
     |   |-- TrainDB
     |       |-- image
     |       |-- scribble
     |   |-- TestDB
     |       |-- CVC-300
     |       |-- CVC-ClinicDB
     |       |-- CVC-ColonDB
     |       |-- ETIS-LaribPolypDB
     |       |-- Kvasir
     

   • Download the pre-trained model.

4. Training:

   • To train our model, run:

   python train_semi_SAM.py

5. Testing:

   • To test our model, just run:

   python test.py

The source code and dataset are free for research and education use only. Any comercial use should get formal permission first.

⬆ back to top