facebookresearch
tutorials for finetuning sam3
Is there any one finetuning sam3 successfully with point and box prompts? Can you share the tutorials for preparing datasets. Thank you very much.
+1. I found the same thing. The current official fine-tuning setup only loads text prompts into the batch, and I couldn't find a configuration to enable box or point prompts. I'm also looking forward to an answer on this.
Perhaps, You can manually provide the input points and input bboxes in COCO_FROM_JSON class of coco_json_loader.py (if you train the model using image.)
See the "loadQueriesAndAnnotationsFromDatapoint" in COCO_FROM_JSON.
Or, if you want to provide the input prompts by randomly sampling them from gt masks, I think you can use the transform class "RandomGeometricInputsAPI" in transforms/point_sampling.py
