Multi-Step Guided Diffusion for Image Restoration on Edge Devices: Toward Lightweight Perception in Embodied AI

TL;DR

This paper introduces a multistep guided diffusion approach that enhances image restoration quality and robustness for embedded AI devices, enabling real-time perception in resource-constrained environments.

Contribution

It proposes a multistep optimization within each denoising step, improving diffusion-based image restoration's fidelity and generalization on edge devices.

Findings

Increasing gradient updates improves LPIPS and PSNR.

Method generalizes well to natural and aerial scenes.

Effective on Jetson Orin Nano with minimal latency.

Abstract

Diffusion models have shown remarkable flexibility for solving inverse problems without task-specific retraining. However, existing approaches such as Manifold Preserving Guided Diffusion (MPGD) apply only a single gradient update per denoising step, limiting restoration fidelity and robustness, especially in embedded or out-of-distribution settings. In this work, we introduce a multistep optimization strategy within each denoising timestep, significantly enhancing image quality, perceptual accuracy, and generalization. Our experiments on super-resolution and Gaussian deblurring demonstrate that increasing the number of gradient updates per step improves LPIPS and PSNR with minimal latency overhead. Notably, we validate this approach on a Jetson Orin Nano using degraded ImageNet and a UAV dataset, showing that MPGD, originally trained on face datasets, generalizes effectively to natural and aerial scenes. Our findings highlight MPGD's potential as a lightweight, plug-and-play restoration module for real-time visual perception in embodied AI agents such as drones and mobile robots.