Noise Flow: Noise Modeling with Conditional Normalizing Flows

TL;DR

Noise Flow introduces a deep learning-based noise model using normalizing flows that accurately captures real sensor noise, outperforming traditional models with fewer parameters.

Contribution

The paper presents Noise Flow, the first deep learning-based noise model that combines parametric noise models with normalizing flows for improved accuracy.

Findings

Outperforms existing noise models with 0.42 nats/pixel improvement

Achieves 52% better likelihood of sampled noise

Uses fewer than 2500 parameters for multiple cameras

Abstract

Modeling and synthesizing image noise is an important aspect in many computer vision applications. The long-standing additive white Gaussian and heteroscedastic (signal-dependent) noise models widely used in the literature provide only a coarse approximation of real sensor noise. This paper introduces Noise Flow, a powerful and accurate noise model based on recent normalizing flow architectures. Noise Flow combines well-established basic parametric noise models (e.g., signal-dependent noise) with the flexibility and expressiveness of normalizing flow networks. The result is a single, comprehensive, compact noise model containing fewer than 2500 parameters yet able to represent multiple cameras and gain factors. Noise Flow dramatically outperforms existing noise models, with 0.42 nats/pixel improvement over the camera-calibrated noise level functions, which translates to 52% improvement in the likelihood of sampled noise. Noise Flow represents the first serious attempt to go beyond simple parametric models to one that leverages the power of deep learning and data-driven noise distributions.