A post-hoc pipeline in which a multimodal language model critiques detector outputs and a CLIP classifier fixes flagged labels improves COCO object detection AP by up to 2.7 points across five detectors.
Relation networks for object detection
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Visual-Linguistic Agent: Towards Collaborative Contextual Object Reasoning
A post-hoc pipeline in which a multimodal language model critiques detector outputs and a CLIP classifier fixes flagged labels improves COCO object detection AP by up to 2.7 points across five detectors.