Gather-Excite: Exploiting Feature Context in Convolutional Neural Networks
arXiv:1810.12348
Abstract
While the use of bottom-up local operators in convolutional neural networks (CNNs) matches well some of the statistics of natural images, it may also prevent such models from capturing contextual long-range feature interactions. In this work, we propose a simple, lightweight approach for better context exploitation in CNNs. We do so by introducing a pair of operators: gather, which efficiently aggregates feature responses from a large spatial extent, and excite, which redistributes the pooled information to local features. The operators are cheap, both in terms of number of added parameters and computational complexity, and can be integrated directly in existing architectures to improve their performance. Experiments on several datasets show that gather-excite can bring benefits comparable to increasing the depth of a CNN at a fraction of the cost. For example, we find ResNet-50 with gather-excite operators is able to outperform its 101-layer counterpart on ImageNet with no additional learnable parameters. We also propose a parametric gather-excite operator pair which yields further performance gains, relate it to the recently-introduced Squeeze-and-Excitation Networks, and analyse the effects of these changes to the CNN feature activation statistics.
arXiv commentNeurIPS 2018
References in corpus (5)
Cited by in corpus (41)
- Attention Mechanisms in Computer Vision: A Survey
- Attentional Local Contrast Networks for Infrared Small Target Detection
- ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases
- CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features
- ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Networks
- Visual Transformers: Token-based Image Representation and Processing for Computer Vision
- Natural Adversarial Examples
- ConTNet: Why not use convolution and transformer at the same time?
- Convolutional Neural Networks with Gated Recurrent Connections
- CMT: Convolutional Neural Networks Meet Vision Transformers
- FSPool: Learning Set Representations with Featurewise Sort Pooling
- Dynamic Neural Networks: A Survey
- FcaNet: Frequency Channel Attention Networks
- An Empirical Evaluation on Robustness and Uncertainty of Regularization Methods
- SPGNet: Semantic Prediction Guidance for Scene Parsing
- Category-Specific CNN for Visual-aware CTR Prediction at JD.com
- Global Second-order Pooling Convolutional Networks
- Learn to cycle: Time-consistent feature discovery for action recognition
- SqueezeSegV3: Spatially-Adaptive Convolution for Efficient Point-Cloud Segmentation
- Squeeze-and-Attention Networks for Semantic Segmentation
- Sampled Training and Node Inheritance for Fast Evolutionary Neural Architecture Search
- MSCFNet: A Lightweight Network With Multi-Scale Context Fusion for Real-Time Semantic Segmentation
- Compact Global Descriptor for Neural Networks
- Gated Channel Transformation for Visual Recognition
- Attentional Feature Fusion
- BiCnet-TKS: Learning Efficient Spatial-Temporal Representation for Video Person Re-Identification
- Dual Contrastive Loss and Attention for GANs
- Restoring Images with Unknown Degradation Factors by Recurrent Use of a Multi-branch Network
- Attention-Guided NIR Image Colorization via Adaptive Fusion of Semantic and Texture Clues
- M2Former: Multi-Scale Patch Selection for Fine-Grained Visual Recognition
- DIANet: Dense-and-Implicit Attention Network
- Neural Function Modules with Sparse Arguments: A Dynamic Approach to Integrating Information across Layers
- Linear Context Transform Block
- Dense xUnit Networks
- Deep Reinforced Attention Learning for Quality-Aware Visual Recognition
- RaftMLP: How Much Can Be Done Without Attention and with Less Spatial Locality?
- Deeper or Wider Networks of Point Clouds with Self-attention?
- ConAM: Confidence Attention Module for Convolutional Neural Networks
- GPCA: A Probabilistic Framework for Gaussian Process Embedded Channel Attention
- Efficient Modelling Across Time of Human Actions and Interactions
- Affinity Graph Supervision for Visual Recognition