Research map / Image generation / Super-resolution
Cascaded Diffusion Models for High Fidelity Image Generation
Jonathan Ho, Chitwan Saharia, William Chan, David J. Fleet, Mohammad Norouzi, Tim Salimans
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 conditioning augmentation to the 16->64 super-resolution model (ablation) | Cascaded diffusion pipeline · ImageNet 64x64 small-scale cascading | FID vs train: 6.02 -> 2.13 | Improved |
| cascade 16->64 with conditioning augmentation instead of no cascading (ablation) | non-cascaded 64x64 baseline · ImageNet 64x64 | FID vs train: 2.35 -> 2.13 | Improved |
| cascade 16->64 without conditioning augmentation (ablation) | non-cascaded 64x64 baseline · ImageNet 64x64 | FID vs train: 2.35 -> 6.02 | Hurt |
| add conditioning augmentation in 32->64 cascade (ablation) | CDM 32->64 · ImageNet 64x64 | FID vs train: 1.71 -> 1.48 | Improved |
| use non-truncated instead of truncated conditioning augmentation (ablation) | CDM 32->64 · ImageNet 64x64 | FID vs train: 1.48 vs 1.48 | No clear effect |
| increase truncation time s too high (1501 vs 501) (ablation) | CDM 32->64 truncated · ImageNet 64x64 | FID vs train: 1.48 -> 1.54 | Hurt |
| add conditioning augmentation when conditioning on ground truth low-res data (ablation) | 32->64 super-resolution · ImageNet 64x64 | FID vs train: 0.76 -> 1.11 | Hurt |
| add Gaussian blur augmentation sigma U(0.4,0.6) to 64->256 model (ablation) | 64->256 super-resolution · ImageNet 256x256 | FID vs train: 7.26 -> 6.18 | Improved |
| use wider blur range sigma U(0.4,0.8) instead of U(0.4,0.6) (ablation) | 64->256 super-resolution · ImageNet 256x256 | FID vs train: 6.18 -> 6.90 | Hurt |
| add class conditioning to 64->256 super-resolution (ablation) | 64->256 super-resolution · ImageNet 256x256 | FID vs train: 6.18 -> 5.75 | Improved |
| increase training batch size from 256 to 1024 (ablation) | 64->256 super-resolution · ImageNet 256x256 | FID vs train: 5.75 -> 5.00 | Improved |
| add random left-right flips to super-resolution training (ablation) | 64->256 super-resolution · ImageNet 256x256 | FID vs train: 5.00 -> 4.88 | Improved |
Closest papers by meaning
- CADS: Unleashing the Diversity of Diffusion Models through Condition-Annealed Sampling (ICLR 2024 · similarity 0.90)
- DeepCache: Accelerating Diffusion Models for Free (CVPR 2024 · similarity 0.89)
- DiffuseHigh: Training-Free Progressive High-Resolution Image Synthesis Through Structure Guidance (AAAI 2025 · similarity 0.89)
- CacheQuant: Comprehensively Accelerated Diffusion Models (CVPR 2025 · similarity 0.89)
- Dual-Rate Diffusion: Accelerating diffusion models with an interleaved heavy-light network (NeurIPS 2026 · similarity 0.88)
- Elucidating the design space of classifier-guided diffusion generation (ICLR 2024 · similarity 0.88)
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.
