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

Abstract

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.

References in corpus (12)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
• The Kinetics Human Action Video Dataset
• Wide Residual Networks
• OCNet: Object Context Network for Scene Parsing
• CCNet: Criss-Cross Attention for Semantic Segmentation
• DetNet: A Backbone network for Object Detection
• Deformable ConvNets v2: More Deformable, Better Results
• SlowFast Networks for Video Recognition
• Bag of Tricks for Image Classification with Convolutional Neural Networks
• A Convolutional Encoder Model for Neural Machine Translation

Cited by in corpus (38)

• Attention Mechanisms in Computer Vision: A Survey
• MMDetection: Open MMLab Detection Toolbox and Benchmark
• MAP-Net: Multi Attending Path Neural Network for Building Footprint Extraction from Remote Sensed Imagery
• You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection
• Polarized Self-Attention: Towards High-quality Pixel-wise Regression
• Efficient Attention: Attention with Linear Complexities
• Analyzing Human-Human Interactions: A Survey
• GA-NET: Global Attention Network for Point Cloud Semantic Segmentation
• Relation-Aware Global Attention for Person Re-identification
• Simple and Lightweight Human Pose Estimation
• Attentive Group Equivariant Convolutional Networks
• SA-Net: Shuffle Attention for Deep Convolutional Neural Networks
• Disentangled Self-Attentive Neural Networks for Click-Through Rate Prediction
• Dynamic GCN: Context-enriched Topology Learning for Skeleton-based Action Recognition
• Identification of Pediatric Respiratory Diseases Using Fine-grained Diagnosis System
• Attention Convolutional Binary Neural Tree for Fine-Grained Visual Categorization
• Gated Convolutional Networks with Hybrid Connectivity for Image Classification
• Group Ensemble: Learning an Ensemble of ConvNets in a single ConvNet
• SCG-Net: Self-Constructing Graph Neural Networks for Semantic Segmentation
• Disentangled Non-Local Neural Networks
• On the Integration of Self-Attention and Convolution
• SiamMan: Siamese Motion-aware Network for Visual Tracking
• GPS-Net: Graph Property Sensing Network for Scene Graph Generation
• Revisiting the Sibling Head in Object Detector
• Data Augmentation for Skin Lesion using Self-Attention based Progressive Generative Adversarial Network
• Dual ResGCN for Balanced Scene GraphGeneration
• Channel Equilibrium Networks for Learning Deep Representation
• FlatteNet: A Simple Versatile Framework for Dense Pixelwise Prediction
• Neural Architecture Search for Lightweight Non-Local Networks
• NTIRE 2020 Challenge on Image Demoireing: Methods and Results
• Improving Monocular Depth Estimation by Leveraging Structural Awareness and Complementary Datasets
• BA^2M: A Batch Aware Attention Module for Image Classification
• Linear Context Transform Block
• HR-RCNN: Hierarchical Relational Reasoning for Object Detection
• UPI-Net: Semantic Contour Detection in Placental Ultrasound
• SERank: Optimize Sequencewise Learning to Rank Using Squeeze-and-Excitation Network
• Efficient Modelling Across Time of Human Actions and Interactions
• Microscopic fine-grained instance classification through deep attention