search

bowang-lab / MedSAM

Segment Anything in Medical Images
https://www.nature.com/articles/s41467-024-44824-z
Apache License 2.0
3k stars 416 forks source link

bbox needed for inference? #284

Closed hnyll closed 3 months ago

hnyll commented 4 months ago

i have a custom 2d dataset of images and gt have trained a model. do i absolutely need bbox for inference?

BalasubramanyamEvani commented 4 months ago

I am not a maintainer of this work. However, based on my understanding of SAM models, you typically need an image and a prompt (such as points, bounding boxes, etc.) which gets encoded by the prompt encoder. This is a core aspect of how they function. Specifically, in the case of MedSAM, I believe they require bounding box prompts.

Just wanted to share my observations. The project authors can probably confirm this.

JunMa11 commented 4 months ago

Hi @hnyll ,

The direct answer is yes.

Thank @BalasubramanyamEvani for your great explanation.