Research map / Image generation / Image restoration
Frequency-Aware Feature Fusion for Dense Image Prediction
Linwei Chen, Ying Fu, Lin Gu, Chenggang Clarence Yan, Tatsuya Harada, Gao Ming Huang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Add FreqFusion feature fusion | SegFormer-B1 · ADE20K val | mIoU: 41.7 -> 44.5 | Improved |
| Use FreqFusion instead of Dysample-S+ | SegFormer-B1 · ADE20K val | mIoU: 43.6 -> 44.5 | Improved |
| Add FreqFusion to Mask2Former | Mask2Former R50 · Cityscapes val | mIoU: 79.4 -> 80.5 | Improved |
| Add FreqFusion to UPerNet, SegFormer and SegNeXt | UPerNet/SegFormer/SegNeXt · Cityscapes val | mIoU: 78.8 -> 79.8; 78.5 -> 80.1; 79.8 -> 80.8 | Improved |
| Add FreqFusion to Mask2Former with large backbones | Mask2Former · ADE20K val | mIoU: 53.9 -> 55.3 (Swin-B); 56.1 -> 56.8 (Swin-L) | Improved |
| Add FreqFusion to SegNeXt-T | SegNeXt-T · COCO-stuff | mIoU: 38.7 -> 40.7 | Improved |
| Use FreqFusion versus Dysample+ | SegNeXt-T · ADE20K val | mIoU: 42.2 -> 43.5 | Improved |
| Add FreqFusion to SegNeXt-T (speed cost) | SegNeXt-T · Cityscapes inference 1024x2048 | FPS: 26.5 -> 23.0 | Hurt |
| Use FreqFusion upsampling in Faster R-CNN | Faster R-CNN R50 · MS-COCO detection | AP: 37.5 -> 39.4 | Improved |
| Use FreqFusion versus Dysample+ with R101 | Faster R-CNN R101 · MS-COCO detection | AP: 40.5 -> 41.0 | Improved |
| Use FreqFusion in Panoptic FPN | Panoptic FPN R50 · MS-COCO panoptic | PQ: 40.2 -> 42.7 | Improved |
| Add ALPF generator in final fusion (ablation) | FreqFusion · Feature analysis (IntraSim) | IntraSim: 0.727 -> 0.799 | Improved |
Closest papers by meaning
- VFM-Adapter: Adapting Visual Foundation Models for Dense Prediction with Dynamic Hybrid Operation Mapping (AAAI 2025 · similarity 0.83)
- Lotus: Diffusion-based Visual Foundation Model for High-quality Dense Prediction (ICLR 2025 · similarity 0.83)
- Frequency-Adaptive Dilated Convolution for Semantic Segmentation (CVPR 2024 · similarity 0.83)
- Frequency Dynamic Convolution for Dense Image Prediction (CVPR 2025 · similarity 0.83)
- Frequency-Aware Density Control via Reparameterization for High-Quality Rendering of 3D Gaussian Splatting (AAAI 2025 · similarity 0.82)
- Balancing Frequencies and Pixels in Flow Matching (NeurIPS 2026 · similarity 0.82)
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.
