Research map / Architectures and efficiency / CNN design
ClcNet: Improving the Efficiency of Convolutional Neural Network Using Channel Local Convolutions
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 |
|---|---|---|---|
| use CLC blocks in place of MobileNet depthwise separable blocks | clcNet-A vs 1.0 MobileNet-224 · ImageNet-1K validation | Mult-Adds: 569M -> 343M; 40% reduction in computation | Helped |
| use CLC blocks in place of MobileNet depthwise separable blocks | clcNet-A vs 1.0 MobileNet-224 · ImageNet-1K validation | Top-1 Acc.: 70.6% -> 70.4% | Hurt |
| use CLC blocks in place of MobileNet depthwise separable blocks | clcNet-B vs 1.0 MobileNet-224 · ImageNet-1K validation | Top-1 Acc.: 70.6% -> 71.6%; 1.0% increase | Helped |
| use CLC blocks instead of ShuffleNet 2× | clcNet-B vs ShuffleNet 2× · ImageNet-1K validation | Mult-Adds: 524M -> 425M; 19% reduction in computation | Helped |
| use CLC blocks instead of ShuffleNet 2× | clcNet-B vs ShuffleNet 2× · ImageNet-1K validation | Top-1 Acc.: 70.9% (v1) -> 71.6%; 0.7% increase | Helped |
| replace all IGC kernels with grouped convolution, removing FCRF (ablation) | clcNet-A · ImageNet-1K validation | Top-1 Acc.: 70.4% -> 67.7% | Hurt |
| replace all IGC kernels with grouped convolution, removing FCRF (ablation) | clcNet-A · ImageNet-1K validation | Top-5 Acc.: 89.5% -> 87.8% | Hurt |
| increase repetitions of late-stage CLC blocks (c or d) (ablation) | clcNet · ImageNet-1K | Top-1 accuracy | Helped |
| increase repetitions of early-stage CLC blocks (a or b) rather than late-stage blocks (ablation) | clcNet · ImageNet-1K | Top-1 accuracy | Mixed |
| use SGD rather than RMSProp (ablation) | clcNet training · ImageNet-1K | Top-1 accuracy | Helped |
| use weight decay 1.0e-4 rather than 4.0e-5 (ablation) | clcNet training · ImageNet-1K | Top-1 accuracy | Helped |
Closest papers by meaning
- Convolutional Neural Networks With Alternately Updated Clique (CVPR 2018 · similarity 0.87)
- More Is Less: A More Complicated Network With Less Inference Complexity (CVPR 2017 · similarity 0.87)
- ChannelNets: Compact and Efficient Convolutional Neural Networks via Channel-Wise Convolutions (NeurIPS 2018 · similarity 0.86)
- Local Binary Convolutional Neural Networks (CVPR 2017 · similarity 0.86)
- MobileODE: An Extra Lightweight Network (NeurIPS 2025 · similarity 0.86)
- CMT: Convolutional Neural Networks Meet Vision Transformers (CVPR 2022 · 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.