search

facebookresearch / segment-anything

The repository provides code for running inference with the SegmentAnything Model (SAM), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.
Apache License 2.0
47.33k stars 5.61k forks source link

Not good performance in medical image segmentation #48

Open weidai00 opened 1 year ago

weidai00 commented 1 year ago

The demo url has a bad performance in medical image segmentation, how could I fine-tune it? could you give some prompt or code?

tifat58 commented 1 year ago

I also tried using models for medical segmentation using the jupyter notebook. They don't seem to be performing well. Will you release any training code or instructions for fine-tune?

qingshi1 commented 1 year ago

The demo url has a bad performance in medical image segmentation, how could I fine-tune it? could you give some prompt or code?

innat commented 1 year ago

Does this dataset (https://ai.facebook.com/datasets/segment-anything/) contains any medical data? Otherwise it doesn't make sense to expect good performance on medical samples. Fine tuning is needed, (but RadImageNet).

emmanuel-contreras commented 1 year ago

I also tried this with single cell data and doesn't appear to be doing well. Probably wasn't trained on it but would be cool to be able to continue training it on some of this other data. image

JunMa11 commented 1 year ago

Hi @weidai00 @tifat58 @qingshi1 @innat @emmanuel-contreras

We provide a step-by-step tutorial on fine-tuning SAM on 2D and 3D medical image datasets. It requires less than 10G GPU memory. Hope that it could be useful.

https://github.com/bowang-lab/MedSAM#model-training-video-tutorial