Describe the feature
I see that for panoptic segmentation the current models have been trained on COCO.Are there any plans on introducing models trained on different datasets (similar to the ones present in mmsegmentation)
Motivation
Currently all models are trained on the benchmark COCO dataset.As such applying those same models on other datasets like ADE (which has a vast vocabulary of common objects .for eg,it has 9000 images just corresponding to the sky label) may result in poor accuracy for panoptic segmentation purposes
A clear and concise description of the motivation of the feature.
Related resourceshttps://github.com/facebookresearch/Mask2Former/blob/main/configs/ade20k/panoptic-segmentation/swin/maskformer2_swin_large_IN21k_384_bs16_160k.yaml
Additional context
Add any other context or screenshots about the feature request here.
If you would like to implement the feature and create a PR, please leave a comment here and that would be much appreciated.
sguttikon
commented
10 months ago
Describe the feature I see that for panoptic segmentation the current models have been trained on COCO.Are there any plans on introducing models trained on different datasets (similar to the ones present in mmsegmentation) Motivation Currently all models are trained on the benchmark COCO dataset.As such applying those same models on other datasets like ADE (which has a vast vocabulary of common objects .for eg,it has 9000 images just corresponding to the sky label) may result in poor accuracy for panoptic segmentation purposes A clear and concise description of the motivation of the feature. Related resources https://github.com/facebookresearch/Mask2Former/blob/main/configs/ade20k/panoptic-segmentation/swin/maskformer2_swin_large_IN21k_384_bs16_160k.yaml
Additional context Add any other context or screenshots about the feature request here. If you would like to implement the feature and create a PR, please leave a comment here and that would be much appreciated.