Research map / Image generation / Text-to-image diffusion
FICGen: Frequency-Inspired Contextual Disentanglement for Layout-driven Degraded Image Generation
Wenzhuang Wang, Yifan Zhao, Mingcan Ma, Ming Liu, Zhonglin Jiang, Yong Chen, Jia Li
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 FPR (frequency perceiver resampler) (ablation) | FICGen with VFEA + ASFA · DIOR (ablation, Faster R-CNN R50) | mAP: 25.6 -> 27.6 | Improved |
| add ASFA (adaptive spatial-frequency aggregation) (ablation) | FICGen with FPR + VFEA · DIOR (ablation, Faster R-CNN R50) | mAP: 25.5 -> 27.6 | Improved |
| add VFEA (visual-frequency enhanced attention) (ablation) | base L2I model · DIOR (ablation, Faster R-CNN R50) | mAP: 15.7 -> 23.2 | Improved |
| add VFEA to FPR + ASFA (ablation) | FPR + ASFA · DIOR (ablation, Faster R-CNN R50) | mAP: 18.5 -> 27.6 | Improved |
| add FPR and ASFA on top of VFEA (ablation) | VFEA only · DIOR (ablation, Faster R-CNN R50) | mAP: 23.2 -> 27.6 | Improved |
| replace CC-Diff by FICGen for layout-to-image generation | CC-Diff · DIOR-H (Faster R-CNN) | mAP: 23.6 -> 27.6 | Improved |
| replace CC-Diff by FICGen | CC-Diff · RUOD (Faster R-CNN) | mAP: 29.7 -> 37.0 | Improved |
| replace CC-Diff by FICGen | CC-Diff · DAWN (Cascade R-CNN) | mAP: 15.7 -> 22.6 | Improved |
| replace CC-Diff by FICGen | CC-Diff · DIOR-H (Faster R-CNN) | FID: 30.88 -> 31.25 | Mixed |
| replace AeroGen by FICGen | AeroGen · DIOR-H (YOLOv8) | mAP: 29.8 -> 31.2 | Improved |
| mix FICGen synthetic data with real data, fine-tuning on real after synthetic pretraining | Faster R-CNN X101 on ExDARK · ExDARK trainability | mAP: 38.6 -> 40.6 | Improved |
| train on pure FICGen synthetic data instead of real data | Faster R-CNN X101 on ExDARK · ExDARK trainability | mAP: 38.6 -> 26.7 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MIGC — MIGC: Multi-Instance Generation Controller for Text-to-Image Synthesis (CVPR 2024) · better on 10 of 10 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GenRec: Knowing Where to Reconstruct and Where to Generate (NeurIPS 2026 · similarity 0.85)
- Disentangling Dual Image References in Frequency Aware Diffusion Models for Personalized Generation (NeurIPS 2026 · similarity 0.84)
- Disentangled 3D Scene Generation with Layout Learning (ICML 2024 · similarity 0.84)
- LookCloser: Frequency-aware Radiance Field for Tiny-Detail Scene (CVPR 2025 · similarity 0.84)
- Move Anything with Layered Scene Diffusion (CVPR 2024 · similarity 0.84)
- MC-DiT: Contextual Enhancement via Clean-to-Clean Reconstruction for Masked Diffusion Models (NeurIPS 2024 · similarity 0.84)
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.
