PocketDVDNet: Realtime Video Denoising for Real Camera Noise

TL;DR

PocketDVDNet is a lightweight, real-time video denoising model that effectively handles realistic camera noise through model compression, domain-specific training, and knowledge distillation, achieving high quality with significantly reduced size.

Contribution

The paper introduces PocketDVDNet, a novel, highly compressed video denoising network that leverages physics-informed noise modeling and domain-adapted distillation for real-time performance.

Findings

Reduces model size by 74%

Achieves real-time processing of 5-frame patches

Improves denoising quality over baseline models

Abstract

Live video denoising under realistic, multi-component sensor noise remains challenging for applications such as autofocus, autonomous driving, and surveillance. We propose PocketDVDNet, a lightweight video denoiser developed using our model compression framework that combines sparsity-guided structured pruning, a physics-informed noise model, and knowledge distillation to achieve high-quality restoration with reduced resource demands. Starting from a reference model, we induce sparsity, apply targeted channel pruning, and retrain a teacher on realistic multi-component noise. The student network learns implicit noise handling, eliminating the need for explicit noise-map inputs. PocketDVDNet reduces the original model size by 74% while improving denoising quality and processing 5-frame patches in real-time. These results demonstrate that aggressive compression, combined with domain-adapted distillation, can reconcile performance and efficiency for practical, real-time video denoising.