Research map / Architectures and efficiency / CNN design
Compressing Convolutional Neural Networks via Factorized Convolutional Filters
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 |
|---|---|---|---|
| jointly learn convolutional filters and binary filter-selection scalars (ablation) | ResNet-56 · CIFAR-10 | accuracy: 93.14% -> 91.92% | Helped |
| fine-tune after pair-channel pruning (ablation) | CNN-FCF · CIFAR-10 | accuracy: 30.17% -> 92.89% | Helped |
| use pair-channel pruning instead of single-channel pruning (ablation) | CNN-FCF · CIFAR-10 | accuracy: 91.44% -> 30.17% | Mixed |
| fine-tune the pruned model (ablation) | CNN-FCF · CIFAR-10 | accuracy: 30.17% -> 92.89% | Helped |
| allocate distinct pruning ratios across layers rather than identical ratios (ablation) | CNN-FCF, ResNet-56 · CIFAR-10 | Top1↓%: 0.42 -> -0.14 | Helped |
| allocate distinct pruning ratios across layers rather than identical ratios (ablation) | CNN-FCF, ResNet-34 · ImageNet | Top1↓%: 0.59 -> -0.03 | Helped |
| use distinct layerwise pruning ratios (ablation) | CNN-FCF, ResNet-34 · ImageNet | FLOPs↓%: 26.96 -> 16.31 | Hurt |
| use pair-channel rather than single-channel pruning (ablation) | CNN-FCF, ResNet-56 · CIFAR-10 | Top1↓%: 0.42 -> -0.24 | Helped |
| use pair-channel rather than single-channel pruning (ablation) | CNN-FCF, ResNet-34 · ImageNet | Top1↓%: 0.59 -> -0.25 | Helped |
| use jointly trained selection scalars instead of ℓ1-norm ranking (ablation) | ResNet-56 · CIFAR-10 | accuracy after fine-tuning: 92.09% -> 92.89% | Helped |
| use jointly trained selection scalars instead of random filter pruning (ablation) | ResNet-56 · CIFAR-10 | accuracy after fine-tuning: 91.14% -> 92.89% | Helped |
| compress ResNet-34 with CNN-FCF at the reported pruning setting | ResNet-34 · ImageNet | Top1↓%: 0.92 -> 0.51 | Helped |
Closest papers by meaning
- RePr: Improved Training of Convolutional Filters (CVPR 2019 · similarity 0.91)
- DPFPS: Dynamic and Progressive Filter Pruning for Compressing Convolutional Neural Networks from Scratch (AAAI 2021 · similarity 0.91)
- Provable Filter Pruning for Efficient Neural Networks (ICLR 2020 · similarity 0.91)
- Auto-Balanced Filter Pruning for Efficient Convolutional Neural Networks (AAAI 2018 · similarity 0.91)
- Learning Filter Pruning Criteria for Deep Convolutional Neural Networks Acceleration (CVPR 2020 · similarity 0.91)
- Entropy Induced Pruning Framework for Convolutional Neural Networks (AAAI 2024 · similarity 0.88)
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.