Research map / Image generation / Fast diffusion sampling
You Only Look One Step: Accelerating Backpropagation in Diffusion Sampling With Gradient Shortcuts
Hongkun Dou, Zeyu Li, Xingyu Jiang, Hongjue Li, Lijun Yang, Wen Yao, Yue Deng
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 |
|---|---|---|---|
| backpropagate one step (SDO) instead of full backpropagation (DOODL) for gradient computation | pixel diffusion model · gradient of x_N, A100 | time (s): 29.96 -> 1.62 (-91.35%) | Improved |
| backpropagate one step (SDO) instead of AdjointDPM | pixel diffusion model · gradient of x_N, A100 | time (s): 18.73 -> 1.62 | Improved |
| backpropagate one step (SDO) instead of DOODL | pixel diffusion model · gradient of x_N, A100 | memory (GB): 11.25 -> 6.12 (-34.68%) | Improved |
| backpropagate one step (SDO) instead of DOODL | latent diffusion model · gradient of x_N, A100 | time (s): 27.20 -> 1.91 (-88.84%) | Improved |
| backpropagate one step (SDO) instead of DOODL | latent diffusion model · gradient of x_N, A100 | memory (GB): 18.59 -> 10.33 (-34.95%) | Improved |
| use SDO gradient instead of full backpropagation for adversarial initial noise | classifier-guided ImageNet diffusion · adversarial sample generation | attack success ratio (%): 51.61 -> 96.32 | Improved |
| use SDO gradient instead of DOODL for adversarial noise | classifier-guided ImageNet diffusion · adversarial sample generation | FID: 70.4 -> 58.2 | Improved |
| use SDO gradient instead of DOODL for text-guided manipulation | pixel diffusion on CelebA-HQ/FFHQ · text-guided image manipulation | CLIP score: 28.76 -> 32.18 | Improved |
| use SDO gradient instead of FlowGrad for text-guided manipulation | pixel diffusion on CelebA-HQ/FFHQ · text-guided image manipulation | LPIPS: 0.142 -> 0.134 | Improved |
| use SDO gradient instead of DOODL for style-guided generation | latent diffusion · style-guided generation | style loss: 357.2 -> 268.1 | Improved |
| use SDO gradient instead of DOODL for style-guided generation | latent diffusion · style-guided generation | CLIP score: 31.12 -> 30.78 | No clear effect |
| use SDO gradient instead of DOODL for aesthetic enhancement | Stable Diffusion v1.4 · aesthetic enhancement | aesthetic score: 9.02 -> 9.13 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- AdjointDPM — AdjointDPM: Adjoint Sensitivity Method for Gradient Backpropagation of Diffusion Probabilistic Models (ICLR 2024) · better on 8 of 9 shared results
- FreeDoM — FreeDoM: Training-Free Energy-Guided Conditional Diffusion Model (ICCV 2023) · better on 3 of 3 shared results
- DiffusionCLIP — DiffusionCLIP: Text-Guided Diffusion Models for Robust Image Manipulation (CVPR 2022) · 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
- One Step Diffusion via Shortcut Models (ICLR 2025 · similarity 0.91)
- Align Your Steps: Optimizing Sampling Schedules in Diffusion Models (ICML 2024 · similarity 0.91)
- Fast constrained sampling in pre-trained diffusion models (NeurIPS 2025 · similarity 0.90)
- Communication-Efficient Diffusion Denoising Parallelization via Reuse-then-Predict Mechanism (NeurIPS 2025 · similarity 0.89)
- TFG: Unified Training-Free Guidance for Diffusion Models (NeurIPS 2024 · similarity 0.89)
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.89)
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.
