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

25 Apr 2019  ·  Yue Cao, Jiarui Xu, Stephen Lin, Fangyun Wei, Han Hu ·

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. In this paper, we take advantage of this finding to create a simplified network based on a query-independent formulation, which maintains the accuracy of NLNet but with significantly less computation. We further observe that this simplified design shares similar structure with Squeeze-Excitation Network (SENet). Hence we unify them into a three-step general framework for global context modeling. Within the general framework, we design a better instantiation, called the global context (GC) block, which is lightweight and can effectively model the global context. The lightweight property allows us to apply it for multiple layers in a backbone network to construct a global context network (GCNet), which generally outperforms both simplified NLNet and SENet on major benchmarks for various recognition tasks. The code and configurations are released at https://github.com/xvjiarui/GCNet.

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Instance Segmentation COCO minival GCNet (ResNeXt-101 + DCN + cascade + GC r16) mask AP 40.9 # 67
Object Detection COCO minival GCNet (ResNeXt-101 + DCN + cascade + GC r16) box AP 47.9 # 87
AP50 66.9 # 31
AP75 52.2 # 25
Object Detection COCO minival GCnet (ResNet-50-FPN, GRoIE) box AP 40.3 # 164
AP50 62.4 # 58
AP75 44 # 70
APS 24.2 # 56
APM 44.4 # 55
APL 52.5 # 66
Object Detection COCO-O GCNet (RX-101-32x4d-DCN) Average mAP 26.0 # 25
Effective Robustness 4.38 # 24
Object Detection COCO test-dev GCNet (ResNeXt-101 + DCN + cascade + GC r4) box mAP 48.4 # 101
AP50 67.6 # 54
AP75 52.7 # 54
Hardware Burden None # 1
Operations per network pass 54.8G # 1
Instance Segmentation COCO test-dev GCNet (ResNeXt-101 + DCN + cascade + GC r16) mask AP 41.5% # 58

Methods