Closed Caewinix closed 1 year ago
Hi @Caewinny ,
We drop the eye vessel segmentation task since we found that it is not suitable for the bounding box-based segmentation methods.
For example, bounding box prompts would be ambiguous in the following setting. The two vessels share the same bounding box.
(credit to https://www.frontiersin.org/articles/10.3389/fendo.2022.972339/full)
I have noticed that on the branch 0.1, the version mentioned that it can segment the DRIVE dataset. However, when I attempt to train that, I got a disappointed result, and even though I use the provided .pth medsam file.
Could you please tell me the detail of training the dataset of the DRIVE? For instance, the preprocessing procedure of such particular task?