Research map / Architectures and efficiency / CNN design
Learning Versatile Filters for Efficient Convolutional Neural Networks
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 |
|---|---|---|---|
| replace sum of secondary filters (Fcn. 3 concatenation instead of Fcn. 4 summation) with concatenated multi-scale outputs, with reduced filter count (ablation) | spatial versatile LeNet · MNIST | accuracy: 99.20% -> 99.15%/99.22% | Helped |
| share bias across secondary filters (ablation) | Versatile-Model (spatial) · MNIST | accuracy: 99.15% -> 99.22% | Helped |
| apply spatial versatile filters (concatenated) to LeNet | LeNet · MNIST | accuracy: 99.20% -> 99.22% | No clear effect |
| increase channel reduction c-c^ from 1 to 2 (ablation) | Versatile v2 LeNet · MNIST | accuracy: 99.18% -> 99.07% | Hurt |
| increase channel stride g from 1 to 2 (ablation) | Versatile v2 LeNet · MNIST | accuracy: 99.18% -> 99.15% | No clear effect |
| apply spatial versatile filters | AlexNet · ImageNet | top-1 error: 42.9% -> 42.1% | Helped |
| apply spatial versatile filters | ResNet-50 · ImageNet | top-5 error: 7.8% -> 7.6% | Helped |
| add channel versatile filters on top (ablation) | Versatile-AlexNet · ImageNet | top-5 error: 19.5% -> 20.7% | Hurt |
| apply channel versatile filters | ResNet-50 · ImageNet | top-5 error: 7.8% -> 8.2% | Hurt |
| apply channel versatile filters | ResNeXt-50 · ImageNet | top-5 error: 6.5% -> 7.0% | Hurt |
| apply channel versatile filters | ShuffleNet 2x (g=3) · ImageNet | top-1 error: 26.3% -> 27.6% | Hurt |
Closest papers by meaning
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.90)
- Compressing Convolutional Neural Networks via Factorized Convolutional Filters (CVPR 2019 · similarity 0.89)
- RePr: Improved Training of Convolutional Filters (CVPR 2019 · similarity 0.89)
- The Master Key Filters Hypothesis: Deep Filters Are General (AAAI 2025 · similarity 0.88)
- Deep Edge Filter: Return of the Human-Crafted Layer in Deep Learning (NeurIPS 2025 · similarity 0.87)
- The Quest for Universal Master Key Filters in DS-CNNs (NeurIPS 2025 · 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.