Questation about the multi-class for the "High Quality Segmentation for Ultra High-resolution Images"

qqlu / Entity

EntitySeg Toolbox: Towards Open-World and High-Quality Image Segmentation

Other

704 stars 58 forks source link

Questation about the multi-class for the "High Quality Segmentation for Ultra High-resolution Images" #18

Open Ianresearch opened 2 years ago

Ianresearch commented 2 years ago

Hello! Have you tried to refine the mask which is more than one class at same time, such as 3 classes or 4 classes? And how about the result? Thank you very much.

tcShen commented 2 years ago

The model is only in class-agnostic style now. Thanks.

Ianresearch commented 2 years ago

Thank you for reply. And I have another queatation.'The perturbed masks are randomly perturbed on ground truth with a random IoU threshold between 0.8 and 1.0' -- can you explain the perturb method more detail,and open the code.Look forward your reply.

tcShen commented 2 years ago

https://github.com/dvlab-research/Entity/blob/main/High-Quality-Segmention/util/boundary_modification.py The perturbation code is in the link above. You can take several minutes to read it. Thanks.

Ianresearch commented 2 years ago

OK. Thank you very much!