Research map / Architectures and efficiency / Attention and transformers
Global Filter Networks for Image Classification
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 self-attention/MLP token mixing with global filter layer (FFT-based) | GFNet-XS vs ResMLP-12 · ImageNet-1k val | top-1 accuracy (%): 76.6 -> 78.6 | Helped |
| global filter instead of self-attention | GFNet-Ti vs DeiT-Ti · ImageNet-1k val | top-1 accuracy (%): 72.2 -> 74.6 | Helped |
| global filter instead of self-attention | GFNet-S vs DeiT-S · ImageNet-1k val | top-1 accuracy (%): 79.8 -> 80.0 | No clear effect |
| hierarchical GFNet-H instead of ResNet/RegNet/PVT | GFNet-H-Ti · ImageNet-1k val | top-1 accuracy (%): 75.1 -> 80.1 vs PVT-Ti | Helped |
| GFNet-H-B instead of Swin-S | GFNet-H-B · ImageNet-1k val | top-1 accuracy (%): 83.0 -> 82.9 | No clear effect |
| fine-tune at 384 resolution for 30 epochs | GFNet-B · ImageNet-1k val | top-1 accuracy (%): 80.7 -> 82.1 | Helped |
| GFNet-S vs DeiT-S under adversarial attack | GFNet-S · FGSM/PGD | accuracy (%): 40.7 -> 42.6 FGSM; 16.7 -> 21.0 PGD | Helped |
| GFNet-S vs DeiT-S on ImageNet-A | GFNet-S · ImageNet-A | top-1 accuracy (%): 18.9 -> 14.3 | Hurt |
| replace global filter with FNet, local depthwise conv, or ResMLP mixing (ablation) | GFNet-XS · ImageNet-1k val | top-1 accuracy (%) | Hurt |
| use single residual connection per block (ablation) | GFNet-XS · ImageNet-1k val | top-1 accuracy (%): +0.2% | Helped |
Closest papers by meaning
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.88)
- LSNet: See Large, Focus Small (CVPR 2025 · similarity 0.87)
- Deep Edge Filter: Return of the Human-Crafted Layer in Deep Learning (NeurIPS 2025 · similarity 0.86)
- The Master Key Filters Hypothesis: Deep Filters Are General (AAAI 2025 · similarity 0.86)
- Enhancing Vision Transformer: Amplifying Non-Linearity in Feedforward Network Module (ICML 2024 · similarity 0.86)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · 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.