RL-Assisted SAM Segmentation in Diverse Medical Imaging Modalities

Code for the paper [Bridging the Gap in Lesion Segmentation: RL-Assisted SAM Segmentation in Diverse Medical Imaging Modalities]

Dependencies

• python 3.6.5
• numpy 1.14.5
• scipy 1.1.0
• Pytorch 0.4.0

Scripts

The folder 'scripts' contains the different bash scripts that could be used to train the same models used in the paper, for both Camvid and Cityscapes datasets.

• launch_baseline.sh: To train the baselines 'random'.
• launch_train_ralis.sh: To train the 'SAM RL agent' model.
• launch_test_ralis.sh: To test the 'SAM RL agent' model.

Pre-trained and Fine-tuned Models

To download the trained pre-trained models for RL: [Google Drive][https://drive.google.com/file/d/15_3dhBL4HmU-U5f3-vJp2lfTFBnTIKLu/view?usp=drive_link]

To download the trained fine-tuned models for SAM_RL: [Google Drive][https://drive.google.com/file/d/1rYKJD_lPScF7JwdGY3NVjqdOO7zzl_LO/view?usp=drive_link]