Difference between ADE20k and SceneParsing150

[aliasvishnu]

Is SceneParsing150 the original snapshot of ADE20k?

Multiple papers refer to ADE20k has having 20210 training images, which matches this description of SceneParsing150. Can someone clarify if this is the case?

This is important if we want to compare our results with this leaderboard. Confirmation will be greatly appreciated.

Thank you.

[hangzhaomit]

Sorry for the confusion. The full ADE20K dataset has 3000+ class annotations, SceneParse150 (150 classes) is a subset of the annotations. But our users usually use ADE20K to refer to SceneParse150. So, if you want to compare with the leaderboard, use SceneParse150!