Residual Objectness replaces hand-crafted sampling and reweighting with cascaded learned objectness refinements, improving RetinaNet, YOLOv3, and Faster R-CNN by 1.1 to 1.3 AP on COCO.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Residual Objectness for Imbalance Reduction
Residual Objectness replaces hand-crafted sampling and reweighting with cascaded learned objectness refinements, improving RetinaNet, YOLOv3, and Faster R-CNN by 1.1 to 1.3 AP on COCO.