README.md

Expected dataset structure for cityscapes:

cityscapes/
  gtFine/
    train/
      aachen/
        *_color.png, *_instanceIds.png, *_labelIds.png, *_polygons.json,
        *_labelTrainIds.png
      ...
    val/
    test/
    cityscapes_panoptic_train_trainId.json
    cityscapes_panoptic_train_trainId/
      *_panoptic.png
    cityscapes_panoptic_val.json
    cityscapes_panoptic_val/
      *_panoptic.png
  leftImg8bit/
    train/
    val/
    test/

Install cityscapes scripts by:

pip install git+https://github.com/mcordts/cityscapesScripts.git

Note:

• We train model with train_id (continuous class label from 0 to 18) and evaluate model with original class label.
• labelTrainIds.png are created by python cityscapesscripts/preparation/createTrainIdLabelImgs.py.
• panoptic.png are created by
  • python cityscapesscripts/preparation/createPanopticImgs.py --use-train-id for generating training labels.
  • python cityscapesscripts/preparation/createPanopticImgs.py for generating evaluation labels.

Expected dataset structure for COCO panoptic segmentation:

coco/
  annotations/
    instances_{train,val}2017.json
    panoptic_{train,val}2017.json
    panoptic_{train,val}2017_trainId.json
    panoptic_{train,val}2017/  # png annotations
  {train,val}2017/
    # image files that are mentioned in the corresponding json

Install panopticapi by:

pip install git+https://github.com/cocodataset/panopticapi.git

Note:

• panoptic_{train,val}2017_trainId.json are created by python prepare_coco_panoptic_trainid.py.