Research map / Architectures and efficiency / CNN design
ELASTIC: Improving CNNs With Dynamic Scaling Policies
Huiyu Wang, Aniruddha Kembhavi, Ali Farhadi, Alan L. Yuille, Mohammad Rastegari
Check 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 Elastic scaling branches | ResNeXt50 · ImageNet validation | Top-1 error: 22.23 -> 21.56 | Improved |
| Add Elastic scaling branches | ResNeXt101 · ImageNet validation | Top-1 error: 21.18 -> 20.83 | Improved |
| Add Elastic scaling branches | DLA-X60 · ImageNet validation | Top-1 error: 21.92 -> 21.25 | Improved |
| Add Elastic scaling branches | DenseNet201 · MSCOCO multi-label classification | F1-PerClass: 69.95 -> 70.40 | Improved |
| Add Elastic scaling branches | ResNeXt50 · MSCOCO multi-label classification | F1-Overall: 74.52 -> 75.37 | Improved |
| Add Elastic scaling branches | ResNeXt50 · MSCOCO small objects | F1-PerClass: 45.57 -> 46.67 | Improved |
| Add Elastic scaling branches | ResNeXt50 · PASCAL VOC val | mIOU: 75.29 -> 77.70 | Improved |
| Use nearest up/downsampling with average pooling instead of bilinear sampling (ablation) | DLA-X60+Elastic · ImageNet validation | Top-1 error: 21.25 -> 21.49 | Hurt |
| Use trained dilated filters with average pooling instead of bilinear sampling (ablation) | DLA-X60+Elastic · ImageNet validation | Top-1 error: 21.25 -> 21.20 | No clear effect |
| Use trained dilated filters without average pooling instead of bilinear sampling (ablation) | DLA-X60+Elastic · ImageNet validation | Top-1 error: 21.25 -> 21.60 | Hurt |
| Use trained filters for up/downsampling instead of bilinear sampling (ablation) | DLA-X60+Elastic · ImageNet validation | Top-1 error: 21.25 -> 21.52 | Hurt |
| Allocate 75% of branches to high resolution instead of splitting equally (ablation) | DLA-X60+Elastic · ImageNet validation | Top-1 error: 21.25 -> 21.35 | Hurt |
Closest papers by meaning
- Deep Elastic Networks With Model Selection for Multi-Task Learning (ICCV 2019 · similarity 0.87)
- Revisiting ResNets: Improved Training and Scaling Strategies (NeurIPS 2021 · similarity 0.87)
- Elastic ViTs from Pretrained Models without Retraining (NeurIPS 2025 · similarity 0.86)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.86)
- Cluster-Learngene: Inheriting Adaptive Clusters for Vision Transformers (NeurIPS 2024 · similarity 0.85)
- Towards flexible perception with visual memory (ICML 2025 · similarity 0.85)
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.
