search

dvlab-research / PanopticFCN

Fully Convolutional Networks for Panoptic Segmentation (CVPR2021 Oral)
Apache License 2.0
391 stars 53 forks source link

The result of cityscapes is strange #49

Open canglangzhige opened 1 year ago

canglangzhige commented 1 year ago

Thanks for your work. When I train the model with Cityscapes follow https://github.com/dvlab-research/PanopticFCN/issues/21. I get the result like:

frankfurt_000000_000576_leftImg8bit

But the ground truth is:

frankfurt_000000_000576_gtFine_panoptic

You can see that the segmentation of the front end at the bottom of the image(inside the red rectangular box) is chaotic. Why is that?

yanwei-li commented 1 year ago

Hi, this is the background class that is not included in the evaluation. It is chaotic because we do not take this class in training, which causes the random prediction. You can take this ignore class into training GT if needed.