Detectron2-Train-a-Instance-Segmentation-Model icon indicating copy to clipboard operation
Detectron2-Train-a-Instance-Segmentation-Model copied to clipboard

Published 20 hours ago • verified publisher icon TannerGilbert

Improving the mask boundary accuracy

Open Rvv1296 opened this issue 2 years ago • 1 comments

I trained a instance segmentation detectron2 model for custom data set. But the mask that I am getting for my detected elements is not perfect. Is there any way to improve the boundary of the mask of instance segmentation result ? Thanks in advance.

Rvv1296 avatar Dec 01 '22 06:12 Rvv1296

@Rvv1296 sorry for the late reply. There are many ways of increasing the performance, including using a bigger dataset, different model architecture as well as some post-processing refinement methods like BPR (Boundary Patch Refinement), which I myself haven't tried yet though.

TannerGilbert avatar Mar 19 '23 09:03 TannerGilbert