Research map / Architectures and efficiency / CNN design
Deep Roots: Improving CNN Efficiency With Hierarchical Filter Groups
Yani Ioannou, Duncan Robertson, Roberto Cipolla, Antonio Criminisi
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Apply hierarchical filter groups to NiN spatial convolutions (root-8) | Network in Network · CIFAR10 | FLOPS: only 46% of the floating point operations | Improved |
| Apply hierarchical filter groups to NiN spatial convolutions (root-8) | Network in Network · CIFAR10 | parameters: 33% of the model parameters | Improved |
| Use the root topology, with grouping decreasing through depth, instead of constant or increasing grouping (ablation) | Network in Network · CIFAR10 | accuracy at a given model size and computational complexity | Improved |
| Apply hierarchical filter groups to spatial convolutions (root-16) | ResNet-50 · ILSVRC | Top-1 accuracy: exceeds the baseline accuracy by 0.2% | Improved |
| Use more filter groups (root-64 rather than the baseline) (ablation) | ResNet-50 · ILSVRC | accuracy: a drop in accuracy of only 0.1% | Hurt |
| Apply hierarchical filter groups to spatial convolutions (root-32) | ResNet-200 · ILSVRC 2012 | FLOPS: 25% fewer FLOPS | Improved |
| Apply hierarchical filter groups to spatial convolutions | GoogLeNet · ILSVRC | FLOPS: 44% fewer floating point operations (multiply-add) | Improved |
| Apply hierarchical filter groups to spatial convolutions | GoogLeNet · ILSVRC | Top-5 accuracy: 0.3% lower top-5 accuracy | Hurt |
Closest papers by meaning
- Inference, Learning and Attention Mechanisms that Exploit and Preserve Sparsity in CNNs (IJCV 2020 · similarity 0.88)
- Growing Efficient Deep Networks by Structured Continuous Sparsification (ICLR 2021 · similarity 0.87)
- ThiNet: Pruning CNN Filters for a Thinner Net (TPAMI 2018 · similarity 0.87)
- RePr: Improved Training of Convolutional Filters (CVPR 2019 · similarity 0.87)
- ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression (ICCV 2017 · similarity 0.87)
- Deep Expander Networks: Efficient Deep Networks from Graph Theory (ECCV 2018 · similarity 0.87)
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.
