Research map / Architectures and efficiency / CNN design
FSNet: Compression of Deep Convolutional Neural Networks by Filter Summary
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 independent convolution filters with overlapping Filter Summary segments | ResNet-110 · CIFAR-10 | accuracy: 93.91% -> 93.81% | No clear effect |
| Add 8-bit weight quantization to FSNet (ablation) | ResNet-110 · CIFAR-10 | parameters: 0.44M -> 0.12M | Helped |
| Add 8-bit weight quantization to FSNet (ablation) | ResNet-164 · CIFAR-10 | accuracy: 94.59% -> 94.65% | Helped |
| Add 8-bit weight quantization to FSNet (ablation) | DenseNet-100 (k=12) · CIFAR-10 | accuracy: 94.46% -> 94.40% | Hurt |
| Add 8-bit weight quantization to FSNet (ablation) | DenseNet-100 (k=24) · CIFAR-10 | accuracy: 95.40% -> 95.43% | Helped |
| Increase convolution compression from r = 2 to r = 3.7 (ablation) | FSNet with ResNet-50 · ImageNet (ILSVRC-12 validation) | Top-1 accuracy: 73.11% -> 70.36% | Hurt |
| Add 8-bit weight quantization to FSNet (ablation) | FSNet-1 · ImageNet (ILSVRC-12 validation) | Top-1 accuracy: 73.11% -> 72.59% | Hurt |
| Add 8-bit weight quantization to FSNet (ablation) | FSNet-1 · VOC 2007 test | mAP: 67.60% -> 67.63% | Helped |
| Add 8-bit weight quantization to FSNet (ablation) | FSNet-2 · VOC 2007 test | mAP: 70.14% -> 70.00% | Hurt |
| Replace DARTS search-space 1 × 1 convolutions with differentiable Filter Summary layers | DARTS · CIFAR-10 | accuracy: 97.50% -> 97.19% | Hurt |
| Increase Filter Summary compression ratio from 4.58 to 13.76 (ablation) | FSNet with ResNet-56 · CIFAR-10 | accuracy: 94.12% -> 92.87% | Hurt |
| Use four cyclical training cycles instead of one baseline-matched cycle (ablation) | FSNet with DenseNet-100 (k=24) · CIFAR-10 | accuracy: 95.46% -> 95.40% | Hurt |
Closest papers by meaning
- Compressing Convolutional Neural Networks via Factorized Convolutional Filters (CVPR 2019 · similarity 0.90)
- Compressing Deep Convolutional Neural Networks by Stacking Low-dimensional Binary Convolution Filters (AAAI 2021 · similarity 0.89)
- CNNpack: Packing Convolutional Neural Networks in the Frequency Domain (NeurIPS 2016 · similarity 0.88)
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.88)
- Deep Edge Filter: Return of the Human-Crafted Layer in Deep Learning (NeurIPS 2025 · similarity 0.86)
- Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks (NeurIPS 2025 · similarity 0.86)
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.