GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond

Yue CaoJiarui XuStephen LinFangyun WeiHan Hu

   Papers with code   Abstract  PDF

The Non-Local Network (NLNet) presents a pioneering approach for capturing long-range dependencies, via aggregating query-specific global context to each query position. However, through a rigorous empirical analysis, we have found that the global contexts modeled by non-local network are almost the same for different query positions within an image... (read more)

Benchmarked Models

RANK
MODEL
REPO
CODE RESULT
PAPER RESULT
ε-REPRODUCED
BUILD
1
GCNet
(ResNeXt-101 + DCN + c…)
0.479
0.479
2
GCNet
(ResNeXt-101 + DCN + c…)
0.478
0.479
3
Mask R-CNN
(ResNeXt-101 + DCN + c…)
0.470
0.471
4
Mask R-CNN
(ResNeXt-101-FPN, sync…)
0.464
0.465
5
Mask R-CNN
(ResNeXt-101-FPN, sync…)
0.458
0.459
6
Mask R-CNN
(ResNeXt-101-FPN, sync…)
0.446
0.447
7
Mask R-CNN
(ResNeXt-101-FPN, sync…)
0.429
0.429
8
Mask R-CNN
(ResNeXt-101-FPN, sync…)
0.424
0.424
9
Mask R-CNN
(ResNet-101-FPN, syncB…)
0.417
0.417
11
Mask R-CNN
(ResNeXt-101-FPN, sync…)
0.415
0.412
12
Mask R-CNN
(ResNet-101-FPN, syncB…)
0.414
0.411
13
Mask R-CNN
(ResNet-50-FPN, GC r4,…)
0.400
0.399
14
Mask R-CNN
(ResNet-50-FPN, GC r16…)
0.397
0.394
15
Mask R-CNN
(ResNet-101-FPN, syncB…)
0.397
0.398
16
Mask R-CNN
(ResNet-101-FPN, fixBN…)
0.397
--
17
Mask R-CNN
(ResNet-101-FPN, 1x LR)
0.397
--
18
Mask R-CNN
(ResNet-50-FPN, 2x LR)
0.382
--
19
Mask R-CNN
(ResNet-50-FPN, 1x LR)
0.373
0.372