RepPoints: Point Set Representation for Object Detection

Ze YangShaohui LiuHan HuLiwei WangStephen Lin

   Papers with code   Abstract  PDF

Modern object detectors rely heavily on rectangular bounding boxes, such as anchors, proposals and the final predictions, to represent objects at various recognition stages. The bounding box is convenient to use but provides only a coarse localization of objects and leads to a correspondingly coarse extraction of object features... (read more)

Benchmarked Models

RANK
MODEL
REPO
CODE RESULT
PAPER RESULT
ε-REPRODUCED
BUILD
1
RPDet
(ResNeXt-101-DCN, mult…)
0.468
0.468
2
RPDet
(ResNeXt-101-FPN-DCN, …)
0.468
--
3
RPDet
(ResNet-101-DCN, multi…)
0.464
0.464
4
RPDet
(ResNet-101-DCN, multi…)
0.464
0.448
5
RPDet
(ResNeXt-101-DCN)
0.445
0.445
6
RPDet
(ResNet-101-FPN, LR 2x…)
0.441
--
7
RPDet
(ResNet-50, multi-scal…)
0.422
0.408
8
RPDet
(ResNet-101-FPN-DCN, L…)
0.416
--
9
RPDet
(ResNet-101)
0.403
0.403
10
RPDet
(ResNet-50)
0.386
0.386