Research map / Image generation / Fast diffusion sampling
ShiftDDPMs: Exploring Conditional Diffusion Models by Shifting Diffusion Trajectories
Zijian Zhang, Zhou Zhao, Jun Yu, Qi Tian
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 Quadratic-Shift instead of conditional reverse-process conditioning | CIFAR-10 image generation · CIFAR-10 | FID: 3.12 -> 3.05 | Improved |
| use Quadratic-Shift instead of conditional reverse-process conditioning | CIFAR-10 image generation · CIFAR-10 | IS: 9.59 -> 9.67 | Improved |
| use Quadratic-Shift instead of conditional reverse-process conditioning | CIFAR-10 image generation · CIFAR-10 | NLL (bits/dim): ≤ 3.74 -> ≤ 3.69 | Improved |
| use Prior-Shift instead of conditional reverse-process conditioning | CIFAR-10 image generation · CIFAR-10 | FID: 3.12 -> 3.06 | Improved |
| use Data-Normalization rather than a conditional DDPM | CIFAR-10 image generation · CIFAR-10 | FID: 3.12 -> 5.51 | Hurt |
| add class labels to the reverse process (ablation) | Quadratic-Shift · CIFAR-10 | IS: 9.67 -> 9.74 | Improved |
| add class labels to the reverse process (ablation) | Quadratic-Shift · CIFAR-10 | FID: 3.05 -> 3.02 | Improved |
| add class labels to the reverse process (ablation) | Quadratic-Shift · CIFAR-10 | NLL (bits/dim): ≤ 3.69 -> ≤ 3.70 | Hurt |
| use Quadratic-Shift for inpainting rather than DDPM (score) | Places2 image inpainting · Places2, 0-20% mask | FID: 2.0665 -> 1.8314 | Improved |
| use Quadratic-Shift for inpainting rather than DDPM (score) | Places2 image inpainting · Places2, 20-40% mask | FID: 6.6129 -> 6.2915 | Improved |
| use Quadratic-Shift for inpainting rather than DDPM (score) | Places2 image inpainting · Places2, 40-60% mask | FID: 17.3601 -> 14.9667 | Improved |
| use Quadratic-Shift instead of conditional DDPM | CUB text-to-image generation · CUB | IS: 4.18 -> 4.42 | Improved |
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.
- StructureFlow — StructureFlow: Image Inpainting via Structure-Aware Appearance Flow (ICCV 2019) · better on 3 of 3 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
- UDPM: Upsampling Diffusion Probabilistic Models (NeurIPS 2024 · similarity 0.91)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.90)
- Diffusion Models Without Attention (CVPR 2024 · similarity 0.90)
- DEFT: Efficient Fine-tuning of Diffusion Models by Learning the Generalised $h$-transform (NeurIPS 2024 · similarity 0.90)
- Alleviating Exposure Bias in Diffusion Models through Sampling with Shifted Time Steps (ICLR 2024 · similarity 0.90)
- D^2-DPM: Dual Denoising for Quantized Diffusion Probabilistic Models (AAAI 2025 · similarity 0.90)
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.
