Research map / Architectures and efficiency / CNN design
URNet: User-Resizable Residual Networks with Conditional Gating Module
Sangho Lee, Simyung Chang, Nojun Kwak
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 |
|---|---|---|---|
| Use input-conditioned residual-block gates trained with scale loss | ResNet-110 · CIFAR-10, scale 0.6 | top-1 accuracy (%): 93.2 -> 93.7 | Improved |
| Use input-conditioned residual-block gates trained with scale loss | ResNet-101 · ImageNet validation, scale 0.8 | top-1 accuracy (%): 76.4 -> 76.4 | No clear effect |
| Use input-conditioned residual-block gates trained with scale loss | Faster-RCNN (ResNet-101) · Pascal VOC 2007 | mAP: 74.96 -> 75.40 | Improved |
| Add binary-gate training alongside sigmoid-gate training (ablation) | URNet-110 · CIFAR-10, scale 0.8 | top-1 accuracy (%): 81.7 -> 92.9 | Improved |
| Add sigmoid-gate training to binary-gate training (ablation) | URNet-110 · CIFAR-10, scale 0.8 | resizability | Improved |
| Replace per-block conditional gates with a separate ResNet-8 gate predictor (ablation) | URNet-110 · CIFAR-10 | scale matching | Hurt |
| Replace hard conditional gates with continuous block attention (ablation) | URNet-110 · CIFAR-10, scale 0.8 | resizability | Hurt |
| Add previous-block features as CGM inputs (ablation) | URNet-110 · CIFAR-10 | gate-usage variance: 0.00 -> 1.03 | Improved |
| Remove previous-block features from CGM inputs (ablation) | URNet-110 · CIFAR-10, scale 0.8 | top-1 accuracy (%): 93.7 -> 93.5 | Hurt |
| Increase the scale-loss weight β too much (ablation) | URNet-110 · CIFAR-10 | top-1 accuracy | Hurt |
| Train at a fixed target scale with scale annealing (ablation) | URNet-101 · ImageNet | top-1 accuracy | Improved |
Closest papers by meaning
- The Reversible Residual Network: Backpropagation Without Storing Activations (NeurIPS 2017 · similarity 0.87)
- Are ResNets Provably Better than Linear Predictors? (NeurIPS 2018 · similarity 0.87)
- Simplified Reversible Residual Networks (NeurIPS 2026 · similarity 0.86)
- Multi-level Residual Networks from Dynamical Systems View (ICLR 2018 · similarity 0.86)
- Residual Alignment: Uncovering the Mechanisms of Residual Networks (NeurIPS 2023 · similarity 0.85)
- Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit (ICLR 2024 · 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.
