Research map / Detection and segmentation / CNN features
Res2Net: A New Multi-Scale Backbone Architecture
Shanghua Gao, Ming‐Ming Cheng, Kai Zhao, Xinyu Zhang, Ming–Hsuan Yang, Philip H. S. Torr
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace bottleneck block with Res2Net module (s=4) | ResNet-50 · ImageNet val | top-1 error (%): 23.85 -> 22.01 | Improved |
| replace block with Res2Net module | ResNeXt-50 · ImageNet val | top-1 error (%): 22.61 -> 21.76 | Improved |
| replace block with Res2Net module | DLA-60 · ImageNet val | top-1 error (%): 23.32 -> 21.53 | Improved |
| add Res2Net module on top of bLResNet | bLResNet-50 · ImageNet val | top-1 error (%): 22.41 -> 21.68 | Improved |
| replace block with Res2Net module in 101-layer network | ResNet-101 · ImageNet val (deeper networks) | top-1 error (%): 22.63 -> 20.81 | Improved |
| increase scale from 2 to 8 at preserved complexity (ablation) | Res2Net-50 · ImageNet val (scale ablation) | top-1 error (%): 22.68 -> 22.01 -> 21.86 | Improved |
| increase scale from 4 to 8 with increased complexity (ablation) | Res2Net-50 · ImageNet val (scale ablation) | top-1 error (%): 22.01 -> 21.42 -> 20.80 | Improved |
| use narrower 18w x 4s Res2Net-50-L (ablation) | Res2Net-50 · ImageNet val (scale ablation) | top-1 error (%): 23.85 -> 22.92 with 2.9G FLOPs vs 4.2G | Improved |
| increase scale beyond 4 on small images (ablation) | Res2NeXt-29 · CIFAR-100 | top-1 error (%) | No clear effect |
| add SE block to Res2NeXt-29 (ablation) | Res2NeXt-29 6c x 24w x 6s · CIFAR-100 | top-1 error (%): 16.79 -> 16.56 | Improved |
| replace backbone with Res2Net-50 | Faster R-CNN · COCO detection | AP (%): 31.1 -> 33.7 | Improved |
| replace backbone with Res2Net-50 | DeepLab v3+ · PASCAL VOC12 val | mean IoU (%): 77.7 -> 79.2 | Improved |
Later papers that compared with this method
1 later accepted paper reports Res2Net-101 as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Aggregation With Feature Detection (ICCV 2021 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Revisiting ResNets: Improved Training and Scaling Strategies (NeurIPS 2021 · similarity 0.91)
- CBNet: A Novel Composite Backbone Network Architecture for Object Detection (AAAI 2020 · similarity 0.88)
- Multi-level Residual Networks from Dynamical Systems View (ICLR 2018 · similarity 0.88)
- Battle of the Backbones: A Large-Scale Comparison of Pretrained Models across Computer Vision Tasks (NeurIPS 2023 · similarity 0.88)
- Residual Alignment: Uncovering the Mechanisms of Residual Networks (NeurIPS 2023 · similarity 0.88)
- Dr2Net: Dynamic Reversible Dual-Residual Networks for Memory-Efficient Finetuning (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
