RegionDet shows supervised detectors can partially learn to localize activity and context regions, while open-vocabulary detectors nearly fail, indicating current vision-language models are object-centric.
Advances in Neural Information Processing Systems , volume=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
RegionDet: A Benchmark for Region Detection Beyond Object Instances
RegionDet shows supervised detectors can partially learn to localize activity and context regions, while open-vocabulary detectors nearly fail, indicating current vision-language models are object-centric.