Unified Video Dense Prediction from Disjoint Data

Scene understanding requires simultaneous prediction about geometry, appearance, and semantics.

Scene understanding requires simultaneous prediction about geometry, appearance, and semantics. However, existing task-specific annotations are fragmented across incompatible, domain-specific datasets. Current unified systems circumvent this by restricting training to fully…