ZeroShotSAM is a project focused on zero-shot medical image segmentation using a sparse prompt approach, leveraging a finetuned version of the Segment Anything Model (SAM). This repository contains code and resources for finetune and evaluating the model, as well as documentation to help users get started.
- Utilizes the Segment Anything Model (SAM) for medical image segmentation.
- Implements zero-shot segmentation using sparse prompts.
- Provides a finetuning mechanism to adapt SAM to medical domain data.
- Includes a script for finetuning and evaluation.
We used the following datasets in our experiments:
CUDA_VISIBLE_DEVICES=0 python -W ignore train.py --task glas --vit vit_b --epoches 100
CUDA_VISIBLE_DEVICES=1 python -W ignore train.py --task glas --vit vit_l --epoches 100
CUDA_VISIBLE_DEVICES=2 python -W ignore train.py --task monu --vit vit_b --epoches 100