search

guobaoxiao / DSAM

Exploring Deeper! Segment Anything Model with Depth Perception for Camouflaged Object Detection, ACM Multimedia (MM), 2024
3 stars 0 forks source link

About Box Prompts #2

Open liujia99 opened 1 month ago

liujia99 commented 1 month ago

Dear author, thank you for your wonderful work.

I have doubts about the box prompts in the code. Is it reasonable that the box information is generated from the GT image either in the training or testing phase?

guobaoxiao commented 1 month ago

Thank you for your interest in our work and for your question. During the training phase, using the box prompt from the GT is equivalent to using the GT for supervision. During the testing phase, we obtain inaccurate boxes from the GT (by randomly shifting the coordinates by 0–20 pixels) to simulate the mode of manual box prompting. This manual box prompt mode is derived from SAM, and DSAM is a promptable segmentation method. Please feel free to reach out if you have any further questions!