Undocumented part label 5 for person and rider (Cityscapes)

[qizhuli]

I downloaded a fresh copy of the dataset from the Cityscapes official website yesterday. While I was processing the data, I found that there seemed to be an undocumented part label 5 for the person and rider class. These are the images that have it:

munster_000033_000019_gtFinePanopticParts.tif: (person, 5)
munster_000035_000019_gtFinePanopticParts.tif: (person, 5)
munster_000046_000019_gtFinePanopticParts.tif: (person, 5)
munster_000047_000019_gtFinePanopticParts.tif: (person, 5), (rider, 5)
munster_000048_000019_gtFinePanopticParts.tif: (person, 5)
munster_000052_000019_gtFinePanopticParts.tif: (person, 5)
munster_000055_000019_gtFinePanopticParts.tif: (person, 5)
munster_000057_000019_gtFinePanopticParts.tif: (person, 5)
munster_000058_000019_gtFinePanopticParts.tif: (person, 5)

where (person, 5) and (rider, 5) means there is at least one pixel with the semantic class person and rider, respectively, and a part label of 5. Upon a quick glance, they seem to correspond to personal accessories and belongings in the images.

I guess it would be better to document them, just to avoid unwanted surprises for dataset users.

[pmeletis]

Any possible ids outside the valid range are automatically handled in v2 release of the code and they are mapped to the ignored class. Thank you for pointing this out!

dspec = pp.DatasetSpec('spec_path.jaml')
outputs_with_guaranteed_range = pp.decode_uids(uids, experimental_dataset_spec=dspec, experimental_correct_range=True)