Res2Net: A New Multi-Scale Backbone Architecture

By Shanghua Gao, Ming-Ming Cheng, Kai Zhao, Xinyu Zhang, Ming-Hsuan Yang, Philip H. S. Torr. 2019

3,066 citations recorded by the source corpus.

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/.

Subfields: Computer Vision and Pattern Recognition, Model Architectures

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Model Architectures

Citations in the corpus

The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization (2020, background): “Notice Res2Net architectures (Gao et al., 2019) can greatly improve accuracy.”
Natural Adversarial Examples (2019, methodology): “We consider the ResNet-50 architecture with its residual blocks exchanged with recently introduced Res2Net v1b blocks [15].”
Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021, positive): “Moreover, with years of rapid developments, there have been many well-engineered CNN backbones such as Res2Net [16], EfficientNet [60], and ResNeSt [77].”
Multiscale Vision Transformers (2021, background): “Incorporating downsampling, shift invariance, and shared weights, ConvNets are de-facto standard backbones for computer vision tasks for image [70, 67, 94, 96, 51, 15, 18, 39, 99, 87, 46] and video [93, 36, 14, 85, 74, 112, 102, 34, 111, 40, 33, 123, 62].”
Strip Pooling: Rethinking Spatial Pooling for Scene Parsing (2020, methodology): “For continuous feature space learning, prior work use multi-scale feature aggregation [37, 5, 33, 18, 42, 31, 32, 2, 44, 4, 48, 17] to fuse the contextual information by probing the incoming features with ﬁlters or pooling operations at multiple rates and multiple ﬁelds-of-view.”
Towards Large-Scale Small Object Detection: Survey and Benchmarks (2022, background): “However, the generic feature extractor [10], [11], [12] usually leverage sub-samplingoperationstoﬁlternoisyactivation[41]andreduce thespatialresolutionoffeaturemaps,thusinevitablylosing theinformationofobjects.”
SwinNet: Swin Transformer Drives Edge-Aware RGB-D and RGB-T Salient Object Detection (2022, neutral): “1) The Effectiveness of Swin Transformer Backbone: We replaces Swin Transformer backbone with some CNN backbones (e.g., ResNet-50 [96], Res2Net-50 [97], ResNet-101 [96], Res50+ViT16 [98]) and transformer backbones (e.g., T2T-14 [66] and PVT-M [68]) to check the effectiveness of backbones.”
3rd Continual Learning Workshop Challenge on Egocentric Category and Instance Level Object Understanding (2022, methodology): “There is a teacher model trained from the previous task, which consists of Res2Net backbone, FPN [23] neck and detector heads.”
LayerCAM: Exploring Hierarchical Class Activation Maps for Localization (2021, methodology): “Researchers have proposed many attention methods [11], [12], [13], [14], [15] to locate the object regions of the class of interest from the powerful CNN based image classiﬁers [16], [17], [18], [19], [20], [21], [22].”
Interactive Knowledge Distillation (2020, background): “ssion. I. INTRODUCTION O VER the past few years, deeper and deeper convolutional neural networks (CNNs) have demonstrated cutting edge performance on various computer vision tasks [1], [2], [3], [4], [5]. However, they usually undergo huge computational costs and memory consumption, and are difﬁcult”
ALL Snow Removed: Single Image Desnowing Algorithm Using Hierarchical Dual-tree Complex Wavelet Representation and Contradict Channel Loss (2021, background): “The two sub-networks leverage the Res2Net [27] as the backbone.”
Rethinking RGB-D Salient Object Detection: Models, Data Sets, and Large-Scale Benchmarks (2019, methodology): “The FPN is independent of the backbone, thus for simplicity, we adopt the VGG-16 [98] architecture as our basic convolutional network to extract spatial features, while utilizing more powerful backbone [99] feacture extractor could be explored in future.”
Delving Deep Into Label Smoothing (2020, neutral): “D EEP Neural Networks (DNNs) [1], [2], [3], [4], [5], [6], [7] have achieved remarkable performance in image recognition [8], [9].”
CBNet: A Composite Backbone Network Architecture for Object Detection (2021, methodology): “, ResNet [11], ResNeXt [13], Res2Net [14], DetNAS [16] and Swin Transformer [17]), and is initialized directly from the pre-trained weights of the single backbone.”
EdgeNeXt: Efficiently Amalgamated CNN-Transformer Architecture for Mobile Vision Applications (2022, methodology): “The first part of our encoder is inspired by Res2Net [12] where we adopt a multi-scale processing approach by developing hierarchical representation into a single block.”
Multi-Scale Spatial Temporal Graph Convolutional Network for Skeleton-Based Action Recognition (2021, background): “The main idea is inspired by Res2Net (Gao et al. 2019) which has a great impact on many ﬁelds of image processing.”