Research map / Architectures and efficiency / CNN design
Understanding the Effective Receptive Field in Deep Convolutional Neural Networks
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 |
|---|---|---|---|
| Stack more layers of random-weight linear convolutions (ERF grows with depth) | deep linear CNNs with random weights · synthetic networks, 5-40 layers | ERF size vs theoretical RF: ERF grows as sqrt(n); relative ratio shrinks as 1/sqrt(n) | Hurt |
| Add ReLU nonlinearity to random-weight CNN | 20-layer random-weight CNN · synthetic networks | ERF Gaussianity | No clear effect |
| Replace 3 of 15 conv layers with stride-2 subsampling or dilated convolutions (factors 2,4,8) (ablation) | 15-layer dense convnet · synthetic networks | effective receptive field size | Helped |
| Train ResNet on CIFAR-10 and CamVid (ERF growth during training) | ResNet classification and segmentation models · CamVid and CIFAR-10 | ERF size: CamVid ERF diameter grows from 100 to around 150 | Helped |
| Use scaled initialization with smaller center and larger outer kernel weights | CNN on CIFAR-10 · CIFAR-10 | training speed: 30% speed-up in a few cases; overall benefit not always significant | Mixed |
Closest papers by meaning
- Receptive Fields As Experts in Convolutional Neural Architectures (ICML 2024 · similarity 0.91)
- Learning Adaptive Receptive Fields for Deep Image Parsing Network (CVPR 2017 · similarity 0.87)
- Structured Receptive Fields in CNNs (CVPR 2016 · similarity 0.86)
- Improving the Effective Receptive Field of Message-Passing Neural Networks (ICML 2025 · similarity 0.86)
- QuARF: Quality-Adaptive Receptive Fields for Degraded Image Perception (AAAI 2025 · similarity 0.84)
- Neural Fields as Distributions: Signal Processing Beyond Euclidean Space (CVPR 2024 · 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.