Wider or Deeper: Revisiting the ResNet Model for Visual Recognition

Abstract

The trend towards increasingly deep neural networks has been driven by a general observation that increasing depth increases the performance of a network. Recently, however, evidence has been amassing that simply increasing depth may not be the best way to increase performance, particularly given other limitations. Investigations into deep residual networks have also suggested that they may not in fact be operating as a single deep network, but rather as an ensemble of many relatively shallow networks. We examine these issues, and in doing so arrive at a new interpretation of the unravelled view of deep residual networks which explains some of the behaviours that have been observed experimentally. As a result, we are able to derive a new, shallower, architecture of residual networks which significantly outperforms much deeper models such as ResNet-200 on the ImageNet classification dataset. We also show that this performance is transferable to other problem domains by developing a semantic segmentation approach which outperforms the state-of-the-art by a remarkable margin on datasets including PASCAL VOC, PASCAL Context, and Cityscapes. The architecture that we propose thus outperforms its comparators, including very deep ResNets, and yet is more efficient in memory use and sometimes also in training time. The code and models are available at https://github.com/itijyou/ademxapp

Code available at: https://github.com/itijyou/ademxapp

References in corpus (21)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Deep Residual Learning for Image Recognition
• Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs
• MXNet: A Flexible and Efficient Machine Learning Library for Heterogeneous Distributed Systems
• Wide Residual Networks
• Multi-Scale Context Aggregation by Dilated Convolutions
• Fully Convolutional Networks for Semantic Segmentation
• DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs
• FractalNet: Ultra-Deep Neural Networks without Residuals
• Learning Deconvolution Network for Semantic Segmentation
• Residual Networks Behave Like Ensembles of Relatively Shallow Networks
• Systematic evaluation of CNN advances on the ImageNet
• Semantic Understanding of Scenes through the ADE20K Dataset
• Semantic Image Segmentation via Deep Parsing Network
• Places: An Image Database for Deep Scene Understanding
• Bridging Category-level and Instance-level Semantic Image Segmentation
• BoxSup: Exploiting Bounding Boxes to Supervise Convolutional Networks for Semantic Segmentation
• Instance-aware Semantic Segmentation via Multi-task Network Cascades
• Exploring Context with Deep Structured models for Semantic Segmentation
• The Loss Surface of Residual Networks: Ensembles and the Role of Batch Normalization
• Dense CNN Learning with Equivalent Mappings

Cited by in corpus (65)

• Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
• The ApolloScape Open Dataset for Autonomous Driving and its Application
• Curriculum Domain Adaptation for Semantic Segmentation of Urban Scenes
• Semantic Instance Segmentation with a Discriminative Loss Function
• CCNet: Criss-Cross Attention for Semantic Segmentation
• Dual Attention Network for Scene Segmentation
• The Reversible Residual Network: Backpropagation Without Storing Activations
• Semantic Understanding of Scenes through the ADE20K Dataset
• Automatic Skin Lesion Analysis using Large-scale Dermoscopy Images and Deep Residual Networks
• STEm-Seg: Spatio-temporal Embeddings for Instance Segmentation in Videos
• Context Encoding for Semantic Segmentation
• Weakly Supervised Medical Diagnosis and Localization from Multiple Resolutions
• Automatic Liver Lesion Detection using Cascaded Deep Residual Networks
• Stacked Deconvolutional Network for Semantic Segmentation
• Learning Pixel-level Semantic Affinity with Image-level Supervision for Weakly Supervised Semantic Segmentation
• Semantic-aware Grad-GAN for Virtual-to-Real Urban Scene Adaption
• A Constructive Prediction of the Generalization Error Across Scales
• A Survey on Deep Learning Methods for Robot Vision
• Constructing Self-motivated Pyramid Curriculums for Cross-Domain Semantic Segmentation: A Non-Adversarial Approach
• Guided Upsampling Network for Real-Time Semantic Segmentation
• Domain Adaptation for Semantic Segmentation with Maximum Squares Loss
• Improving Semantic Segmentation via Self-Training
• MaskLab: Instance Segmentation by Refining Object Detection with Semantic and Direction Features
• Vortex Pooling: Improving Context Representation in Semantic Segmentation
• GFF: Gated Fully Fusion for Semantic Segmentation
• Stacked U-Nets: A No-Frills Approach to Natural Image Segmentation
• Sharing Residual Units Through Collective Tensor Factorization in Deep Neural Networks
• A Good Practice Towards Top Performance of Face Recognition: Transferred Deep Feature Fusion
• Scaling Laws for Deep Learning
• ORDNet: Capturing Omni-Range Dependencies for Scene Parsing
• Knowledge Adaptation for Efficient Semantic Segmentation
• Multi-View Deep Learning for Consistent Semantic Mapping with RGB-D Cameras
• Regularizing Proxies with Multi-Adversarial Training for Unsupervised Domain-Adaptive Semantic Segmentation
• Semantic Flow for Fast and Accurate Scene Parsing
• Penalizing Top Performers: Conservative Loss for Semantic Segmentation Adaptation
• Adaptive Masked Proxies for Few-Shot Segmentation
• Box-driven Class-wise Region Masking and Filling Rate Guided Loss for Weakly Supervised Semantic Segmentation
• Advancing System Performance with Redundancy: From Biological to Artificial Designs
• Efficient Segmentation: Learning Downsampling Near Semantic Boundaries
• Tree-structured Kronecker Convolutional Network for Semantic Segmentation
• Decoders Matter for Semantic Segmentation: Data-Dependent Decoding Enables Flexible Feature Aggregation
• A Curriculum Domain Adaptation Approach to the Semantic Segmentation of Urban Scenes
• Learning to Predict Context-adaptive Convolution for Semantic Segmentation
• Learning Strict Identity Mappings in Deep Residual Networks
• A Multiscale Patch Based Convolutional Network for Brain Tumor Segmentation
• Dynamic-structured Semantic Propagation Network
• Sparsely Aggregated Convolutional Networks
• White matter hyperintensity segmentation from T1 and FLAIR images using fully convolutional neural networks enhanced with residual connections
• A Network Structure to Explicitly Reduce Confusion Errors in Semantic Segmentation
• Monocular Depth Estimation with Augmented Ordinal Depth Relationships
• Part-level Car Parsing and Reconstruction from Single Street View
• ShelfNet for Fast Semantic Segmentation
• Rethinking Convolutional Semantic Segmentation Learning
• EfficientFCN: Holistically-guided Decoding for Semantic Segmentation
• How to make a pizza: Learning a compositional layer-based GAN model
• Adaptive Context Network for Scene Parsing
• Diagnostics in Semantic Segmentation
• Tandem Blocks in Deep Convolutional Neural Networks
• Unsupervised Video Object Segmentation with Distractor-Aware Online Adaptation
• Design Pseudo Ground Truth with Motion Cue for Unsupervised Video Object Segmentation
• A Holistically-Guided Decoder for Deep Representation Learning with Applications to Semantic Segmentation and Object Detection
• Learning Propagation for Arbitrarily-structured Data
• Sequential image processing methods for improving semantic video segmentation algorithms
• Exploring and Exploiting Diversity for Image Segmentation
• 3D Scene Parsing via Class-Wise Adaptation