Abstract: Current scene graph datasets suffer from strong long-tail distributions of their predicate classes. Due to a very low number of some predicate classes in the test sets, no reliable metrics ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback