search

facebookresearch / Mask2Former

Code release for "Masked-attention Mask Transformer for Universal Image Segmentation"
MIT License
2.48k stars 379 forks source link

Semantic segmentation on Coco #80

Open EnnioEvo opened 2 years ago

EnnioEvo commented 2 years ago

I can't see any configs or pretrained weights related to the Coco dataset. I want to train semantic segmentation on my Coco + custom data with mask2former how can I do that. As an alternative, I have already trained panoptic segmentation, but how can I get semantic inference from the panoptic training? The reason I want semantic segmentation is that there are too many unlabelled areas.

PS. thanks for this amazing job, panoptic works great on my dataset

bowenc0221 commented 2 years ago

I believe the current panoptic segmentation model returns semantic segmentation output. Please check the last row of your metrics, which should be mIoU.

EnnioEvo commented 2 years ago

I can't report the mIoU because the evaluation doesn't work (but the panoptic training works great) The output are a set of masks that don't cover the whole picture Or is there another way to run the demo for the semantic segmentation in another way?