I transformed the segmentation map of Cityscapes Validation Datasets (frankfurt, lindau, and munster) into Mask images (by changing the value of the road, sidewalks... to 255 and the rest to 0.)
Then, we have 500 pictures in total to evaluate ADVENT.

The JSON files are at /network/tmp1/ccai/data/mask_generation/Cityscapes/structuredData/val
and the exact images are at /network/tmp1/ccai/data/Cityscapes/structuredData/val

I used it to validate ADVENT model. The mIOU is between 83.23 and 88.32. Because the "ground truth" is the union of road, sidewalks and anything on the ground rather than "where to flood", the mIOU can only be a reference. If you want to know the detailed performance, please checks the result here:
https://www.comet.ml/tianyu-z/advent-ccai/e6eb250447ac4d668c8e2f3225a7fd91?experiment-tab=images