Research map / Architectures and efficiency / CNN design
Learning in the Frequency Domain
Check 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 spatial RGB downsampling with frequency-domain DCT input | ResNet-50 · ImageNet validation | Top-1 accuracy: 75.780 -> 77.194 | Helped |
| Replace spatial RGB input with YCbCr input | ResNet-50 · ImageNet validation | Top-1 accuracy: 75.780 -> 75.234 | Hurt |
| Use 24 selected DCT channels instead of RGB input | ResNet-50 · ImageNet validation | Top-1 accuracy: 75.780 -> 77.196 | Helped |
| Use 24 channels with square, triangle, or dynamic selection patterns (ablation) | ResNet-50 · ImageNet validation | Top-1 accuracy: 77.166; 77.196; 77.148 | No clear effect |
| Increase selected channels from 24 to 48 or 64 (ablation) | ResNet-50 DCT · ImageNet validation | Top-1 accuracy: 77.196 -> 77.384; 77.196 -> 77.232 | Mixed |
| Use triangle rather than square channel selection (ablation) | ResNet-50 DCT · ImageNet validation | Top-1 accuracy: 77.196 -> 77.148; 77.384 -> 77.338; 77.232 -> 77.280 | Mixed |
| Use DCT channels instead of RGB input | MobileNetV2 · ImageNet validation | Top-1 accuracy: 71.702 -> 71.776; 72.156; 72.364; 72.282 | Helped |
| Increase selected channels from 6 to 12 (ablation) | MobileNetV2 DCT · ImageNet validation | Top-1 accuracy: 71.776 -> 72.156 | Helped |
| Increase selected channels from 12 to 24 or 32 (ablation) | MobileNetV2 DCT · ImageNet validation | Top-1 accuracy: 72.156 -> 72.364; 72.156 -> 72.282 | Mixed |
| Use frequency-domain input with 24 selected channels | Mask R-CNN ResNet-50-FPN · COCO 2017 validation | bbox AP: 37.3 -> 37.7 | Helped |
| Use frequency-domain input with 24 selected channels | Mask R-CNN ResNet-50-FPN · COCO 2017 validation | mask AP: 34.2 -> 34.6 | Helped |
| Increase selected channels from 24 to 48 or 64 (ablation) | Mask R-CNN DCT · COCO 2017 validation | bbox AP: 37.7 -> 38.1 | Helped |
Closest papers by meaning
- Frequency-Domain Dynamic Pruning for Convolutional Neural Networks (NeurIPS 2018 · similarity 0.87)
- Beyond Spatial Domain: Cross-domain Promoted Fourier Convolution Helps Single Image Dehazing (AAAI 2025 · similarity 0.87)
- Multiscale Low-Frequency Memory Network for Improved Feature Extraction in Convolutional Neural Networks (AAAI 2024 · similarity 0.86)
- Frequency-Aware Deepfake Detection: Improving Generalizability through Frequency Space Domain Learning (AAAI 2024 · similarity 0.86)
- Addressing Spectral Bias of Deep Neural Networks by Multi-Grade Deep Learning (NeurIPS 2024 · similarity 0.86)
- Frequency-Adaptive Dilated Convolution for Semantic Segmentation (CVPR 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.