ReSWD: ReSTIR'd, not shaken. Combining Reservoir Sampling and Sliced Wasserstein Distance for Variance Reduction

2025/10/01 by Mark Boss, Andreas Engelhardt, Boss, Mark +5
Earth and Planetary Sciences · Engineering · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Graphics (cs.GR) #Hydraulic Fracturing and Reservoir Analysis #Machine Learning (cs.LG) #Reservoir Engineering and Simulation Methods #Seismic Imaging and Inversion Techniques

paper · pdf · doi:10.48550/arxiv.2510.01061

openalex publication_date 2025/10/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Distribution matching is central to many vision and graphics tasks, where the widely used Wasserstein distance is too costly to compute for high dimensional distributions. The Sliced Wasserstein Distance (SWD) offers a scalable alternative, yet its Monte Carlo estimator suffers from high variance, resulting in noisy gradients and slow convergence. We introduce Reservoir SWD (ReSWD), which integrates Weighted Reservoir Sampling into SWD to adaptively retain informative projection directions in optimization steps, resulting in stable gradients while remaining unbiased. Experiments on synthetic benchmarks and real-world tasks such as color correction and diffusion guidance show that ReSWD consistently outperforms standard SWD and other variance reduction baselines. Project page: https://reservoirswd.github.io/

Citations

• Color Transfer with Modulated Flows
• Color Conditional Generation with Sliced Wasserstein Guidance
• Multiscale Sliced Wasserstein Distances as Perceptual Color Difference Measures
• Long-Time Asymptotics of the Sliced-Wasserstein Flow
• Sliced-Wasserstein Distances and Flows on Cartan-Hadamard Manifolds
• Stereographic Spherical Sliced Wasserstein Distances
• Sliced Wasserstein with Random-Path Projecting Directions
• Geometry and analytic properties of the sliced Wasserstein space
• Quasi-Monte Carlo for 3D Sliced Wasserstein
• Sin3DM: Learning a Diffusion Model from a Single 3D Textured Shape
• Sliced Wasserstein Estimation with Control Variates
• Energy-Based Sliced Wasserstein Distance
• Adding Conditional Control to Text-to-Image Diffusion Models
• Adding Conditional Control to Text-to-Image Diffusion Models
• DreamFusion: Text-to-3D using 2D Diffusion
• Exploring CLIP for Assessing the Look and Feel of Images
• Generating natural images with direct Patch Distributions Matching
• High-Resolution Image Synthesis with Latent Diffusion Models
• High-Resolution Image Synthesis with Latent Diffusion Models
• PhotoWCT²: Compact Autoencoder for Photorealistic Style Transfer Resulting from Blockwise Training and Skip Connections of High-Frequency Residuals
• Domain-Aware Universal Style Transfer
• LoRA: Low-Rank Adaptation of Large Language Models
• Learning Transferable Visual Models From Natural Language Supervision
• Spatiotemporal reservoir resampling for real-time ray tracing with dynamic direct lighting
• Distributional Sliced-Wasserstein and Applications to Generative Modeling
• Ultrafast Photorealistic Style Transfer via Neural Architecture Search
• Max-Sliced Wasserstein Distance and its use for GANs
• Photorealistic Style Transfer via Wavelet Transforms
• Generalized Sliced Wasserstein Distances
• Interpolating between Optimal Transport and MMD using Sinkhorn Divergences
• A Closed-form Solution to Photorealistic Image Stylization
• Universal Style Transfer via Feature Transforms
• Deep Photo Style Transfer
• Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization
• Color transfer between images

Related

• Sliced Wasserstein Estimation with Control Variates
• Markovian Sliced Wasserstein Distances: Beyond Independent Projections
• Sliced Wasserstein with Random-Path Projecting Directions
• Energy-Based Sliced Wasserstein Distance
• Efficient Sliced Wasserstein Distance Computation via Adaptive Bayesian Optimization