Fine-tuning SAM and MicroSAM for Microscopy Image Segmentation

Overview

This project focuses on fine-tuning Segment Anything Model (SAM) and MicroSAM for microscopy image segmentation. The repository provides a structured approach to utilizing pretrained SAM, fine-tuning SAM on a custom labeled dataset, applying watershed-based segmentation to post-process fine-tuned SAM results, and implementing MicroSAM as an alternative to compare segmentation performance on a custom image dataset.

Files

The following files are included in this repository:

1. pretrained_SAM.ipynb - demonstrates the segmentaiton performance of pretrained SAM for microscopy image.
2. finetuned_SAM.ipynb - details the process of fine-tuning SAM model on custom microscopy images to improve segmentation performance.
3. watershed.ipynb - implements watershed-based segmentation, to converts the probability map from the result of fine-tuned SAM to multiple masks segmentation.
4. MicroSAM.ipynb - applied MicroSAM, an optimized version of SAM for microscopy image segmentation.