Code repository supporting the paper "Auto-Generating Weak Labels for Real & Synthetic Data to Improve Label-Scarce Medical Image Segmentation" - MIDL 2024
In the original MedSAM paper, there is no clear indication that point prompts are directly supported as input to MedSAM. They just emphasize the use of bounding box prompts.
It appears that in this paper, you have devised a method to generate point prompts for MedSAM based on the coarse labels from the initial model, even though the original MedSAM paper does not explicitly mention supporting point prompts.
Can you please elaborate a little more on this? How is it possible?
In the original MedSAM paper, there is no clear indication that point prompts are directly supported as input to MedSAM. They just emphasize the use of bounding box prompts. It appears that in this paper, you have devised a method to generate point prompts for MedSAM based on the coarse labels from the initial model, even though the original MedSAM paper does not explicitly mention supporting point prompts. Can you please elaborate a little more on this? How is it possible?