Research map / Architectures and efficiency / CNN design
Dynamic Resolution Network
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| dynamic resolution selection vs fixed 224 ResNet-50 | ResNet-50 · ImageNet-100 | top-1 accuracy: 78.5 -> 82.5 | Helped |
| dynamic resolution with stronger FLOPs regularization (alpha=2.5) | ResNet-50 · ImageNet-100 | top-1 accuracy: 78.5 -> 81.4 | Helped |
| replace resolution-aware BN with a shared BN (ablation) | DRNet ResNet-50 · ImageNet-100 | top-1 accuracy: 82.5 -> 80.3 | Hurt |
| increase target FLOPs alpha from 2.0 to 3.5 (ablation) | DRNet ResNet-50 · ImageNet-100 | top-1 accuracy: 80.6 -> 82.6 | Helped |
| increase penalty eta from 0.1 to 1 (ablation) | DRNet ResNet-50 · ImageNet-100 | top-1 accuracy: 82.1 -> 81.5 | Hurt |
| extend candidate resolutions to [224,192,168,112,96] (ablation) | DRNet ResNet-50 · ImageNet-100 | top-1 accuracy: 82.5 -> 81.9 | No clear effect |
| dynamic resolution vs baseline ResNet-50 | ResNet-50 · ImageNet-1K | top-1 accuracy: 76.1 -> 77.5 | Helped |
| dynamic resolution vs baseline ResNet-101 | ResNet-101 · ImageNet-1K | top-1 accuracy: 77.4 -> 79.0 | Helped |
| lower alpha from 3.5 to 2.0 (ablation) | DR-ResNet-50 · ImageNet-1K | top-1 accuracy: 77.4 -> 75.3 | Hurt |
| dynamic resolution vs static 192x192 input | ResNet-50 · ImageNet-1K | top-1 accuracy: 74.3 -> 77.5 | Helped |
| dynamic resolution vs baseline MobileNetV2 | MobileNetV2 · ImageNet-1K | top-1 accuracy: 71.8 -> 72.7 | Helped |
| dynamic resolution predictor vs random resolution selection (ablation) | DR-ResNet-50 · ImageNet-1K | accuracy | Helped |
Closest papers by meaning
- Unified Dynamic Convolutional Network for Super-Resolution With Variational Degradations (CVPR 2020 · similarity 0.88)
- Image Super-Resolution via Deep Recursive Residual Network (CVPR 2017 · similarity 0.87)
- Deeply-Recursive Convolutional Network for Image Super-Resolution (CVPR 2016 · similarity 0.87)
- Resolution Adaptive Networks for Efficient Inference (CVPR 2020 · similarity 0.87)
- Image Super-Resolution via Dual-State Recurrent Networks (CVPR 2018 · similarity 0.86)
- Glance and Focus: a Dynamic Approach to Reducing Spatial Redundancy in Image Classification (NeurIPS 2020 · similarity 0.86)
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.