Research map / Architectures and efficiency / CNN design
Pooling Revisited: Your Receptive Field Is Suboptimal
Check 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 |
|---|---|---|---|
| Replace fixed resizing with learned DynOPool scales | VGG-16, ResNet-50, and MobileNetV2 · FGVC-Aircraft, CIFAR-100, and ImageNet | accuracy: 85.3 -> 87.4; 75.4 -> 79.8; 73.9 -> 74.1; 81.6 -> 87.2; 78.5 -> 80.6; 77.2 -> 78.1; 77.6 -> 82.6; 73.8 -> 76.2; 71.7 -> | Helped |
| Use DynOPool-S resizing at similar computational cost | VGG-16, ResNet-50, and MobileNetV2 · FGVC-Aircraft, CIFAR-100, and ImageNet | accuracy: 85.3 -> 87.0; 75.4 -> 75.5; 73.9 -> 73.8; 81.6 -> 82.3; 78.5 -> 80.3; 77.2 -> 77.6; 77.6 -> 78.7; 73.8 -> 74.0; 71.7 -> | Mixed |
| Replace Shape Adaptor resizing with DynOPool | VGG-16, ResNet-50, and MobileNetV2 · CIFAR-100 | accuracy: 79.2 -> 79.8; 80.3 -> 80.6; 75.7 -> 76.2 | Helped |
| Add DynOPool resizing | EfficientNet-B0 · ImageNet | accuracy: 71.8 -> 72.3 | Helped |
| Replace HRNet-W48 strided downsampling with DynOPool and stride-1 convolution | HRNet-W48 · PascalVOC | mIoU: 76.2 -> 76.4 | Helped |
Closest papers by meaning
- Detail-Preserving Pooling in Deep Networks (CVPR 2018 · similarity 0.87)
- Recombinator Networks: Learning Coarse-To-Fine Feature Aggregation (CVPR 2016 · similarity 0.86)
- Beyond Skip Connection: Pooling and Unpooling Design for Elimination Singularities (AAAI 2025 · similarity 0.84)
- Receptive Fields As Experts in Convolutional Neural Architectures (ICML 2024 · similarity 0.83)
- HyperNet Fields: Efficiently Training Hypernetworks without Ground Truth by Learning Weight Trajectories (CVPR 2025 · similarity 0.83)
- Pool Me Wisely: On the Effect of Pooling in Transformer-Based Models (NeurIPS 2025 · similarity 0.83)
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.