Fine-tune on static images

[ImanGandomi]

Hi. I have a project for segmenting surgical instruments in cataract surgery. Currently, I don,t have an annotated video dataset. I just have a static annotated image segmentation dataset, which has about 6000 annotated images.
I want to know if I can rain MedSAM2 on my static image segmentation dataset?
Is it logical?

[TangentSnow]

Yes, just don't use the video predictor. There are two methods "build_sam2" and "build_sam2_video_predictor". Just us the regular version to fine-tune on your dataset. Although I think MedSAM2 is not the right model for your use case. The encoder is trained to extract meaningful features for images like CT scans, MRIs, and etc., I don't think it would do a good job extracting meaningful features for surgical instruments. Maybe just regular SAM2 would be better for your use case?