Research map / Image generation / Autoregressive and latent generation
Frido: Feature Pyramid Diffusion for Complex Scene Image Synthesis
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 |
|---|---|---|---|
| Use multi-scale coarse-to-fine diffusion rather than single-scale latent diffusion | Frido on COCO layout-to-image · COCO layout-to-image | FID: 40.91 -> 37.14 | Helped |
| Apply test-time CLIP ranking | Frido-f16f8-G on COCO 2014 T2I · COCO 2014 T2I | FID: 11.24 -> 8.97 | Helped |
| Use a shared U-Net across feature levels (ablation) | Sequential two-LDM baseline on COCO T2I · COCO T2I model analysis | parameters: 1.179 B -> 589.8 M | Helped |
| Use a shared U-Net across feature levels (ablation) | Sequential two-LDM baseline on COCO T2I · COCO T2I model analysis | inference time (sec/img): 1.02865 -> 0.75052 | Helped |
| Add coarse-to-fine modulation (ablation) | Frido PyU-Net on COCO T2I · COCO T2I | generation quality | Helped |
| Use higher-resolution f8f4 features instead of f16f8 (ablation) | Frido on COCO T2I · COCO T2I | FID: 40.14 -> 38.68 | Helped |
| Use higher-resolution f8f4 features instead of f16f8 (ablation) | Frido on COCO T2I · COCO T2I | inference time: 0.918 -> 3.857 | Hurt |
| Add the f32 feature scale (ablation) | Frido on COCO T2I · COCO T2I | FID: 40.14 -> 48.04 | Hurt |
| Add the f32 feature scale (ablation) | MS-VQGAN · COCO-stuff reconstruction | rFID: 4.67 -> 6.68 | Hurt |
| Use no noise augmentation on higher-level conditioning features (ablation) | Frido on COCO T2I · COCO T2I | FID: 106 to 122 | Hurt |
| Set the noise ratio too high (e.g., α=0.5) (ablation) | Frido coarse-to-fine modulation · COCO T2I and Visual Genome SG2I | generation performance | Hurt |
| Use classifier-free guidance | Frido-f16f8 on COCO 2014 T2I · COCO 2014 T2I | FID: 15.38 -> 11.24 | Helped |
Closest papers by meaning
- Patched Denoising Diffusion Models For High-Resolution Image Synthesis (ICLR 2024 · similarity 0.88)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.87)
- Any-Size-Diffusion: Toward Efficient Text-Driven Synthesis for Any-Size HD Images (AAAI 2024 · similarity 0.87)
- Image Inpainting via Tractable Steering of Diffusion Models (ICLR 2024 · similarity 0.87)
- Improving Progressive Generation with Decomposable Flow Matching (NeurIPS 2025 · similarity 0.87)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.87)
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.