Research map / Architectures and efficiency / CNN design
Beyond Filters: Compact Feature Map for Portable Deep Model
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 |
|---|---|---|---|
| fine-tune the reconstructed compact network (ablation) | RedCNN · MNIST LeNet | accuracy (%): 78 -> 98.9 | Helped |
| use RedCNN compression instead of training the same small architecture from scratch (ablation) | RedCNN · MNIST LeNet | accuracy (%): 92.8 -> 99.16 | Helped |
| increase sparsity weight beta (ablation) | RedCNN · MNIST LeNet | accuracy vs compression | Mixed |
| choose beta=0.06 (ablation) | RedCNN · MNIST LeNet | compression/speed-up/accuracy: 11.3×, 8.7×, 99.16 | Helped |
| RedCNN vs original AlexNet | RedCNN · ILSVRC-2012 AlexNet | top-1 error: 41.8% -> 42.1% | Hurt |
| RedCNN vs original VGGNet-16 | RedCNN · ILSVRC-2012 VGGNet-16 | top-1 error: 28.5% -> 29.3% | Hurt |
| RedCNN vs original ResNet-50 | RedCNN · ILSVRC-2012 ResNet-50 | top-1 error: 24.6% -> 25.7% | Hurt |
| RedCNN vs CNNpack on VGGNet-16 speed-up | RedCNN · ILSVRC-2012 VGGNet-16 | speed-up ratio: 9.4× -> 9.63× | Helped |
| RedCNN vs CNNpack on AlexNet speed-up | RedCNN · ILSVRC-2012 AlexNet | speed-up ratio: 25× -> 4.31× | Hurt |
| compress feature maps (RedCNN) rather than filters only | RedCNN · ILSVRC-2012 AlexNet | feature map compression ratio: 1× -> 2.45× | Helped |
| RedCNN filter compression vs pruning and CNNpack on VGGNet-16 | RedCNN · ILSVRC-2012 VGGNet-16 | filter compression ratio: 49× -> 6.87× | Hurt |
| add 8-bit quantization to RedCNN (not run) | RedCNN · ILSVRC-2012 ResNet-50 | filter compression ratio: 4.35× -> 17.4× | Helped |
Closest papers by meaning
- Learning Versatile Filters for Efficient Convolutional Neural Networks (NeurIPS 2018 · similarity 0.88)
- HRank: Filter Pruning Using High-Rank Feature Map (CVPR 2020 · similarity 0.86)
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.86)
- Modulated Convolutional Networks (CVPR 2018 · similarity 0.85)
- Deep Pyramidal Residual Networks (CVPR 2017 · similarity 0.85)
- The Master Key Filters Hypothesis: Deep Filters Are General (AAAI 2025 · similarity 0.84)
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.