Spatially Adaptive Noise Injection
Abstract
Diffusion samplers reverse a learned noising process using either stochastic (DDPM) or deterministic (DDIM) updates, which represent endpoints of a single family controlled by a scalar noise-injection variance that is applied identically at every spatial location. This uniform approach neglects the geometry of natural images: high-curvature regions such as edges and textures, where the denoiser is uncertain, benefit from stochastic correction, whereas smooth regions, where the score is precise, are degraded by injected noise. This work investigates whether each pixel requires stochastic correction at a given timestep and introduces Spatially Adaptive Noise Injection (SANI), a novel sampling framework that dynamically adjusts noise application on a per-pixel basis. SANI integrates a probabilistic gating mechanism with a derived spatially adaptive variance, ensuring that noise is injected precisely where needed to refine complex features while preserving well-formed structures. Experimental results and decoupling ablations demonstrate that SANI consistently improves Fréchet Inception Distance (FID) over the vanilla DDPM and DDIM endpoint samplers across diverse sampling timesteps, while remaining competitive with variance-learning baselines, highlighting the importance of spatial adaptivity in diffusion sampling.
Related
- Training More Robust Classification Model via Discriminative Loss and Gaussian Noise Injection
- Spatially Coupled MacKay-Neal Codes on Symmetric Markov Noise: A Reduction to BMS Channels
- Performance of Planar XYZ and XZZX Codes under Homogeneous and Spatially Nonuniform Noise
- Local Linearization and Exact Oscillations for the Nonlinear Stochastic Heat Equation with Spatially Rough Noise
- LithoSAGE: Spatially Adaptive and Geometrically Equivariant Neural Operators for Computational Lithography
BibTeX
@misc{lavda2026spatially,
title = {{Spatially Adaptive Noise Injection}},
author = {Frantzeska Lavda and Maciej Falkiewicz and Van Khoa Nguyen and Alexandros Kalousis},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.18466},
eprint = {2609.18466},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.18466},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Spatially Adaptive Noise Injection
AU - Frantzeska Lavda
AU - Maciej Falkiewicz
AU - Van Khoa Nguyen
AU - Alexandros Kalousis
PY - 2026
DA - 2026/09/16
PB - arXiv
DO - 10.48550/arxiv.2609.18466
UR - https://arxiv.org/abs/2609.18466
AB - Diffusion samplers reverse a learned noising process using either stochastic (DDPM) or deterministic (DDIM) updates, which represent endpoints of a single family controlled by a scalar noise-injection variance that is applied identically at every spatial location. This uniform approach neglects the geometry of natural images: high-curvature regions such as edges and textures, where the denoiser is uncertain, benefit from stochastic correction, whereas smooth regions, where the score is precise, are degraded by injected noise. This work investigates whether each pixel requires stochastic correction at a given timestep and introduces Spatially Adaptive Noise Injection (SANI), a novel sampling framework that dynamically adjusts noise application on a per-pixel basis. SANI integrates a probabilistic gating mechanism with a derived spatially adaptive variance, ensuring that noise is injected precisely where needed to refine complex features while preserving well-formed structures. Experimental results and decoupling ablations demonstrate that SANI consistently improves Fréchet Inception Distance (FID) over the vanilla DDPM and DDIM endpoint samplers across diverse sampling timesteps, while remaining competitive with variance-learning baselines, highlighting the importance of spatial adaptivity in diffusion sampling.
ER -