Research map / Architectures and efficiency / CNN design
Active Convolution: Learning the Shape of Convolution for Image Classification
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 3×3 convolutions with learnable-position ACUs | Plain network · CIFAR-10 test | error rate: 8.01 -> 7.33 | Helped |
| Replace 3×3 convolutions with learnable-position ACUs | Plain network · CIFAR-100 test | error rate: 27.85 -> 27.11 | Helped |
| Replace residual-block convolutions with ACUs | Basic residual network · CIFAR-10 test | error rate: 8.01 -> 7.54 | Helped |
| Replace residual-block convolutions with ACUs | Basic residual network · CIFAR-100 test | error rate: 30.06 -> 29.38 | Helped |
| Replace the bottleneck 3×3 convolution with an ACU | Bottleneck residual network · CIFAR-10 test | error rate: 7.64 -> 7.12 | Helped |
| Replace the bottleneck 3×3 convolution with an ACU | Bottleneck residual network · CIFAR-100 test | error rate: 27.93 -> 27.47 | Helped |
| Replace 3×3 convolutions with ACUs | AlexNet · Place365-Standard validation | top-1 accuracy: 51.68 -> 52.28 | Helped |
| Replace 3×3 convolutions with ACUs | AlexNet · Place365-Standard validation | top-5 accuracy: 81.29 -> 82.08 | Helped |
| Replace residual-block 3×3 convolutions with ACUs | ResNet26 · Place365-Standard validation | top-1 accuracy: 55.33 -> 55.71 | Helped |
| Replace residual-block 3×3 convolutions with ACUs | ResNet26 · Place365-Standard validation | top-5 accuracy: 85.24 -> 85.73 | Helped |
| Warm up with fixed synapse positions before learning positions and weights jointly (ablation) | ACU networks · Benchmark data | classification performance: 0.1%--0.2% | Helped |
| Normalize position gradients to use their direction rather than magnitude (ablation) | ACU networks · Image classification experiments | training stability and result quality | Helped |
Closest papers by meaning
- Active Learning for Convolutional Neural Networks: A Core-Set Approach (ICLR 2018 · similarity 0.91)
- Involution: Inverting the Inherence of Convolution for Visual Recognition (CVPR 2021 · similarity 0.90)
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.88)
- SCNN: A General Distribution Based Statistical Convolutional Neural Network with Application to Video Object Detection (AAAI 2019 · similarity 0.88)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.86)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · 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.