bowenc0221 / panoptic-deeplab

This is Pytorch re-implementation of our CVPR 2020 paper "Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation" (https://arxiv.org/abs/1911.10194)
Apache License 2.0
587 stars 117 forks source link

about 'unlabeled' category in Detectron2 #84

Closed mvdelt closed 3 years ago

mvdelt commented 3 years ago

in Detectron2 builtin_meta.py, CITYSCAPES_CATEGORIES doesn't contain the 'unlabeled' category. only foreground categories are registered.

So, IMHO, when inference, there should be some 'threshold' to determine the background, such as ROI_HEADS.SCORE_THRESH_TEST, isn't it?

If there is no threshold, the background would be covered with other categories.

How can we set the threshold?

bowenc0221 commented 3 years ago

There is no background class in Cityscapes, all 'unlabeled' categories are ignored in evaluation. Please check the Citysacpes official evaluation code for more detail.

mvdelt commented 3 years ago

Checked the evaluation code, you're right. Thanks!