Res2Net: A New Multi-Scale Backbone Architecture

2019/04/30 by Shang-Hua Gao, Shanghua Gao, Ming-Ming Cheng +8 · 3 citations
Computer Science · Mathematics · #Advanced Image and Video Retrieval Techniques #Advanced Neural Network Applications #Algorithm #Artificial intelligence #Backbone network #Block (permutation group theory) #Computer science #Convolutional neural network #Layer (electronics) #Mathematics #Object detection #Pattern recognition (psychology) #Representation (politics) #Residual #Salient #Scale (ratio) #Visual Attention and Saliency Detection #cs.CV

paper · pdf · doi:10.1109/tpami.2019.2938758

published as IEEE TPAMI 2021 · 11 pages, 7 figures

openalex publication_date 2019/08/30 · arxiv created 2021/01/27 · arxiv updated 2021/01/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/06

Abstract

Representing features at multiple scales is of great importance for numerous vision tasks. Recent advances in backbone convolutional neural networks (CNNs) continually demonstrate stronger multi-scale representation ability, leading to consistent performance gains on a wide range of applications. However, most existing methods represent the multi-scale features in a layer-wise manner. In this paper, we propose a novel building block for CNNs, namely Res2Net, by constructing hierarchical residual-like connections within one single residual block. The Res2Net represents multi-scale features at a granular level and increases the range of receptive fields for each network layer. The proposed Res2Net block can be plugged into the state-of-the-art backbone CNN models, e.g., ResNet, ResNeXt, and DLA. We evaluate the Res2Net block on all these models and demonstrate consistent performance gains over baseline models on widely-used datasets, e.g., CIFAR-100 and ImageNet. Further ablation studies and experimental results on representative computer vision tasks, i.e., object detection, class activation mapping, and salient object detection, further verify the superiority of the Res2Net over the state-of-the-art baseline methods. The source code and trained models are available on https://mmcheng.net/res2net/.

Citations

• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• Very Deep Convolutional Networks for Large-Scale Image Recognition
• The Pascal Visual Object Classes (VOC) Challenge
• ImageNet Large Scale Visual Recognition Challenge
• Distinctive Image Features from Scale-Invariant Keypoints
• Two-Stream Convolutional Networks for Action Recognition in Videos
• ImageNet classification with deep convolutional neural networks
• DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• Rethinking Atrous Convolution for Semantic Image Segmentation
• A Survey of Model Compression and Acceleration for Deep Neural Networks
• Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks with Octave Convolution
• High-Resolution Representations for Labeling Pixels and Regions
• Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection
• Learning Multiple Layers of Features from Tiny Images
• Fully Convolutional Networks for Semantic Segmentation
• Squeeze-and-Excitation Networks

Cited by

• VeLU: Variance-enhanced Learning Unit for Deep Neural Networks
• Beyond Residual Connections: Manifold-Constrained Hyper-Connections for Robust Speaker Representation Learning
• PraNet-V2: Dual-Supervised Reverse Attention for Medical Image Segmentation
• SwinNet: Swin Transformer drives edge-aware RGB-D and RGB-T salient object detection