Research map / Architectures and efficiency / CNN design
SSBNet: Improving Visual Recognition Efficiency by Adaptive Sampling
Ho Man Kwan, Shenghui Song
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 |
|---|---|---|---|
| add adaptive instead of uniform sampling (ablation) | SSB-ResNet-D-152 (16,8,4), RandAugment · ImageNet | Top-1 accuracy: 79.71 -> 80.12 | Improved |
| add adaptive instead of uniform sampling (ablation) | SSB-ResNet-D-101 (16,8,4), no RandAugment · ImageNet | Top-1 accuracy: 78.68 -> 78.88 | Improved |
| add adaptive instead of uniform sampling (ablation) | SSB-ResNet-D-50 (16,8,4), RandAugment · ImageNet | Top-1 accuracy: 78.07 -> 78.22 | No clear effect |
| add adaptive instead of uniform sampling (ablation) | SSB-ResNet-D-50 (12,6,3), RandAugment · ImageNet | Top-1 accuracy: 77.97 -> 77.96 | No clear effect |
| use adaptive sampling instead of uniform bilinear sampling (ablation) | SSB-ResNet-D-152 (16,8,4) · ImageNet | Top-1 accuracy: 78.86 -> 79.05 | Improved |
| replace ResNet-D with SSB (adaptive sampling bottleneck) | ResNet-D-50 · ImageNet | Top-1 accuracy: 78.1 -> 78.1 | No clear effect |
| replace ResNet-D with SSB (adaptive sampling bottleneck) | ResNet-D-152 · ImageNet | Top-1 accuracy: 80.1 -> 79.2 | Hurt |
| replace ResNet-RS with SSB | ResNet-RS-200 · ImageNet | Top-1 accuracy: 83.4 -> 82.6 | Hurt |
| replace EfficientNet with SSB | EfficientNet-B0 · ImageNet | Top-1 accuracy: 76.4 -> 75.4 | Hurt |
| replace ResNet-D-50 with SSB-ResNet-D-50 backbone | Mask R-CNN FPN · COCO-2017 (12 epochs) | AP box: 40.14 -> 40.67 | Improved |
| replace ResNet-D-50 with SSB-ResNet-D-50 backbone | Mask R-CNN FPN · COCO-2017 (36 epochs) | AP box: 42.50 -> 42.89 | Improved |
| replace ResNet-RS-200 with SSB version | ResNet-RS-200 · Latency V100 | latency: 988 -> 744 | Improved |
Closest papers by meaning
- Resolution Adaptive Networks for Efficient Inference (CVPR 2020 · similarity 0.85)
- Adaptive Neural Networks for Efficient Inference (ICML 2017 · similarity 0.85)
- Is Second-Order Information Helpful for Large-Scale Visual Recognition? (ICCV 2017 · similarity 0.85)
- SCNN: A General Distribution Based Statistical Convolutional Neural Network with Application to Video Object Detection (AAAI 2019 · similarity 0.84)
- Adapting Shortcut With Normalizing Flow: An Efficient Tuning Framework for Visual Recognition (CVPR 2023 · similarity 0.84)
- Superpixel Sampling Networks (ECCV 2018 · similarity 0.84)
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.
