I’m working on a machine learning project aimed at testing water purity using microscopic images. The goal of the project is to:

Segment various bacteria present in the sample images.
Recognize different types of bacteria.
Assess the purity of water based on the quantity and types of bacteria identified.

I am having difficulty finding and preparing a suitable dataset of microscopic images of water samples containing different types of bacteria.

Given the specificity of my task, what approach should I take to fine-tune a pretrained model like SAM (Segment Anything Model) for bacterial segmentation? Any tips on hyperparameters, data augmentation, or other training strategies would be helpful.