search

yformer / EfficientSAM

EfficientSAM: Leveraged Masked Image Pretraining for Efficient Segment Anything
Apache License 2.0
2.08k stars 149 forks source link

Finetuning for downstream segmentation task #32

Open agraham9966 opened 8 months ago

agraham9966 commented 8 months ago

Hi there - I noticed the new notebooks added for segment anything example. I am wondering if you can give any advice for how to finetune it on a custom dataset for semantic segmentation? Will there be another notebook released which will cover this?

Thanks

balakv504 commented 8 months ago

Great question! We can try to provide a code for finetuning once we finish address some of the ongoing issues (providing examples for multiple objects, onnx export, etc). Few questions: How many images are there in your custom dataset? If it is small (O(few thousand)), you can try to instantiate the model (as is done in the example notebook) to start the FT process.

yformer commented 8 months ago

@agraham9966, Yes, we added the segment anything example. Do you mean finetuning it for segment anything as SAM, or semantic anything as mask2former?