Research map / Architectures and efficiency / CNN design
DCFNet: Deep Neural Network with Decomposed Convolutional Filters
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 full convolution filters with 14 Fourier-Bessel bases | Conv-2 · MNIST test | accuracy: 99.40 -> 99.47 | No clear effect |
| Replace full convolution filters with 14 Fourier-Bessel bases | Conv-3 · SVHN test | accuracy: 94.22 -> 94.63 | Helped |
| Replace full convolution filters with 14 Fourier-Bessel bases | Conv-3 · CIFAR10 test | accuracy: 85.66 -> 85.88 | Helped |
| Replace full filters with 3 Fourier-Bessel bases | VGG-16 · CIFAR10 test | accuracy: 87.02 -> 88.21 | Helped |
| Use 3 rather than 14 Fourier-Bessel bases (ablation) | Conv-2 · MNIST test | accuracy: 99.47 -> 99.40 | No clear effect |
| Use 3 rather than 14 Fourier-Bessel bases (ablation) | Conv-3 · SVHN test | accuracy: 94.63 -> 92.84 | Hurt |
| Use 3 rather than 14 Fourier-Bessel bases (ablation) | Conv-3 · CIFAR10 test | accuracy: 85.88 -> 83.12 | Hurt |
| Use Fourier-Bessel filters instead of full CNN filters | Face verification network · LFW | accuracy: 97.65 % -> 97.32 % | No clear effect |
| Use Fourier-Bessel decompositions rather than full convolution filters | Conv-3 denoiser · SVHN test, Gaussian noise | PSNR: 30.01 -> 31.24 | Helped |
| Use Fourier-Bessel decompositions rather than full convolution filters | Conv-3 denoiser · SVHN test, Speckle noise | PSNR: 28.15 -> 29.84 | Helped |
| Use Fourier-Bessel bases rather than Gaussian random bases (ablation) | DCFNet classification models · MNIST, SVHN, and CIFAR10 | accuracy | Mixed |
| Use Fourier-Bessel bases rather than PCA bases trained on a small subset (ablation) | DCFNet classification models · SVHN and CIFAR10 | accuracy | Mixed |
Closest papers by meaning
- Doubly Convolutional Neural Networks (NeurIPS 2016 · similarity 0.88)
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.87)
- Decoupled Dynamic Filter Networks (CVPR 2021 · similarity 0.87)
- Compressing Convolutional Neural Networks via Factorized Convolutional Filters (CVPR 2019 · similarity 0.87)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.85)
- Deep Learning as Neural Low-Degree Filtering: A Spectral Theory of Hierarchical Feature Learning (NeurIPS 2026 · similarity 0.85)
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.