An end-to-end transformer couples detection with a temporal matching penalty and feedback queries, boosting temporal consistency of scene-graph predictions on Action Genome, OpenPVSG, and MEVA.
Simple online and realtime tracking
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Temporally Consistent Dynamic Scene Graphs: An End-to-End Approach for Action Tracklet Generation
An end-to-end transformer couples detection with a temporal matching penalty and feedback queries, boosting temporal consistency of scene-graph predictions on Action Genome, OpenPVSG, and MEVA.