Research map / Probabilistic and scientific modeling / Variational and generative sampling
ReSWD: ReSTIR‘d, not shaken. Combining Reservoir Sampling and Sliced Wasserstein Distance for Variance Reduction.
Mark Boss, Andreas Engelhardt, Simon Donné, Varun Jampani
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 ReSWD instead of standard SWD in the color matching pipeline (ablation) | CDL color matching · Color matching (10 scenes) | Color PSNR: 24.30 -> 24.64 | Improved |
| add ReSWD instead of standard SWD in the color matching pipeline (ablation) | CDL color matching · Color matching (10 scenes) | CDL error (RMSE): 0.11 -> 0.10 | Improved |
| add ReSWD instead of standard SWD in the color matching pipeline (ablation) | CDL color matching · Color matching (10 scenes) | Transform error (RMSE): 0.34 -> 0.31 | Improved |
| add ReSWD instead of standard SWD in the color matching pipeline (ablation) | CDL color matching · Color matching (10 scenes) | Time per match (s): 5 -> 5 | No clear effect |
| replace prior SDXL guidance with ReSWD modifications on the same SDXL base | color diffusion guidance · Diffusion guidance (ContraStyles/Unsplash) | Mean-W2 [10^2]: 1.941 -> 1.213 | Improved |
| replace prior SDXL guidance with ReSWD modifications on the same SDXL base | color diffusion guidance · Diffusion guidance (ContraStyles/Unsplash) | Time per generation (s): 124 -> 34 | Improved |
| upgrade base model from SDXL to SD3.5-medium with ReSWD | ReSWD diffusion guidance · Diffusion guidance (ContraStyles/Unsplash) | CLIP-IQA: 0.696 -> 0.786 | Improved |
| use distilled SD3.5-turbo instead of SD3.5-medium | ReSWD diffusion guidance · Diffusion guidance (ContraStyles/Unsplash) | Mean-W2 [10^2]: 0.815 -> 0.675 | Improved |
| use SD3.5-large instead of SD3.5-medium | ReSWD diffusion guidance · Diffusion guidance (ContraStyles/Unsplash) | Mean-W2 [10^2]: 0.815 -> 0.55 | Improved |
| use SD3.5-large instead of SD3.5-medium | ReSWD diffusion guidance · Diffusion guidance (ContraStyles/Unsplash) | Time per generation (s): 32 -> 69 | Hurt |
| set number of new candidates M to 8 out of 64 total projections (ablation) | ReSWD general distribution matching · General distribution matching (d=3) | mean W1 | Improved |
| use too few or too many new candidates per step (ablation) | ReSWD general distribution matching · General distribution matching (d=3) | mean W1 | Hurt |
Closest papers by meaning
- Sliced Wasserstein Estimation with Control Variates (ICLR 2024 · similarity 0.87)
- Sliced Wasserstein with Random-Path Projecting Directions (ICML 2024 · similarity 0.86)
- Quasi-Monte Carlo for 3D Sliced Wasserstein (ICLR 2024 · similarity 0.85)
- Stereographic Spherical Sliced Wasserstein Distances (ICML 2024 · similarity 0.84)
- Hierarchical Hybrid Sliced Wasserstein: A Scalable Metric for Heterogeneous Joint Distributions (NeurIPS 2024 · similarity 0.84)
- Tree-Sliced Wasserstein Distance with Nonlinear Projection (ICML 2025 · similarity 0.83)
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.
