Research map / Image generation / Image restoration
Freqformer: Image-Demoiréing Transformer via Effective Frequency Decomposition
Xiaoyang Liu, Bolin Qiu, Zheng Chen, Libo Zhu, Zihan Zhou, Kai Liu, Jiezhang Cao, Yulun Zhang
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 |
|---|---|---|---|
| use the proposed frequency decomposition instead of Haar DWT (ablation) | Freqformer · FHDMi | PSNR: 23.01 -> 25.26 | Improved |
| use the proposed frequency decomposition instead of Block DCT (ablation) | Freqformer · FHDMi | PSNR: 23.22 -> 25.26 | Improved |
| resize the low-frequency input instead of cropping (ablation) | Freqformer · FHDMi low branch | PSNR: 23.13 -> 29.90 | Improved |
| add crop-and-tile inference to crop training (ablation) | Freqformer · FHDMi low branch | PSNR: 23.13 -> 25.90 | Improved |
| use learnable FCT instead of fixed composition (ablation) | Freqformer · FHDMi | PSNR: 25.13 -> 25.26 | Improved |
| add frequency decomposition with dual branch (vs no decomposition) (ablation) | Freqformer · FHDMi | PSNR: 23.92 -> 25.26 | Improved |
| add hierarchical fusion to high branch (ablation) | Freqformer · FHDMi high branch | PSNR: 29.15 -> 29.20 | Improved |
| use full 3x3 convolution instead of SCP (ablation) | Freqformer · FHDMi high branch | PSNR: 29.20 -> 29.18 | No clear effect |
| replace SA-CA with DAT block (ablation) | Freqformer · FHDMi high branch | PSNR: 29.20 -> 29.10 | Hurt |
| apply Freqformer (6.065M) vs ESDNet-L (10.623M) on UHDM | Freqformer · UHDM | PSNR: 22.42 -> 22.24 | Hurt |
| use Freqformer vs ESDNet-L at 4K inference | Freqformer · 4K speed | s/image: 0.25 -> 0.58 | Hurt |
Closest papers by meaning
- Efficient RAW Image Deblurring with Adaptive Frequency Modulation (NeurIPS 2025 · similarity 0.87)
- UniDemoiré: Towards Universal Image Demoiréing with Data Generation and Synthesis (AAAI 2025 · similarity 0.87)
- PixelART: Image-to-Layer Decomposition without Latents or Text-to-Image Pretraining (NeurIPS 2026 · similarity 0.86)
- FreCaS: Efficient Higher-Resolution Image Generation via Frequency-aware Cascaded Sampling (ICLR 2025 · similarity 0.86)
- FDS: Frequency-Aware Denoising Score for Text-Guided Latent Diffusion Image Editing (CVPR 2025 · similarity 0.85)
- Learning Image Demoiréing from Unpaired Real Data (AAAI 2024 · 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.
