Research map / Image generation / Fast diffusion sampling
Rex: A Family of Reversible Exponential (Stochastic) Runge-Kutta Solvers
Zander W. Blasingame, Chen Liu
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 |
|---|---|---|---|
| use Rex (Dopri5) instead of plain Dopri5 for likelihood computation | DiT on tri-alanine · Tri-alanine Boltzmann sampling | E-W2: 0.737 -> 0.495 | Improved |
| use Rex (Dopri5) instead of plain Dopri5 | DiT on tri-alanine · Tri-alanine Boltzmann sampling | ESS: 0.140 -> 0.104 | Hurt |
| use Rex (Dopri5) instead of plain Dopri5 | DiT on tri-alanine · Tri-alanine Boltzmann sampling | T-W2: 0.468 -> 0.497 | Hurt |
| use Rex (Dopri5) instead of DDIM for round-trip editing | Stable Diffusion v1.5 · pix2pix image editing | LPIPS: 0.214 -> 0.107 | Improved |
| use BDIA reversible solver for editing (lacks non-zero stability region) | Stable Diffusion v1.5 · pix2pix image editing | LPIPS: 0.885 | Hurt |
| use EDICT for image editing | Stable Diffusion v1.5 · pix2pix image editing | edit quality | Hurt |
| use Rex variants instead of other reversible solvers | Stable Diffusion v1.5 · COCO captions text-to-image (1000) | CLIP, Image Reward, PickScore | Improved |
| use stochastic Rex variants (Euler-Maruyama, ShARK) over ODE Rex | Stable Diffusion v1.5 · COCO captions text-to-image (1000) | Image Reward, PickScore | Improved |
| use Rex (Euler) instead of other reversible solvers for round-trip inversion | Stable Diffusion v1.5 · pix2pix latent reconstruction (FP32) | latent MSE | Improved |
| use Rex over EDICT, BDIA and O-BELM for unconditional generation | DDPM on CelebA-HQ 256 · CelebA-HQ unconditional generation | FD and other metrics | Improved |
| use higher-order Rex (RK4) instead of Rex (Euler) (ablation) | DDPM on CelebA-HQ 256 · CelebA-HQ unconditional generation | sample quality | Hurt |
| use SDE Rex at small step counts (ablation) | DDPM on CelebA-HQ 256 · CelebA-HQ unconditional generation | sample quality | Hurt |
Closest papers by meaning
- Adaptive Stochastic Coefficients for Accelerating Diffusion Sampling (NeurIPS 2025 · similarity 0.88)
- Learning to Solve Generative ODEs Beyond the Linear Span (NeurIPS 2026 · similarity 0.87)
- Bespoke Solvers for Generative Flow Models (ICLR 2024 · similarity 0.87)
- Beyond First-Order Tweedie: Solving Inverse Problems using Latent Diffusion (CVPR 2024 · similarity 0.87)
- Lightning-Fast Image Inversion and Editing for Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.87)
- Semantic Image Inversion and Editing using Rectified Stochastic Differential Equations (ICLR 2025 · similarity 0.86)
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.
