Dynamic Pre-training: Towards Efficient and Scalable All-in-One Image Restoration

TL;DR

This paper introduces DyNet, a flexible, efficient all-in-one image restoration network with dynamic pre-training, capable of handling multiple degradations with reduced computational costs and storage, validated on a new large-scale dataset.

Contribution

We propose DyNet, a dynamic encoder-decoder network with shared weights for all-in-one image restoration, and a dynamic pre-training strategy that reduces GPU hours and storage overhead.

Findings

Achieves state-of-the-art results in denoising, deraining, and dehazing.

Reduces model complexity by over 50% in GFlops and parameters.

Demonstrates effective multi-task restoration with a unified model.

Abstract

All-in-one image restoration tackles different types of degradations with a unified model instead of having task-specific, non-generic models for each degradation. The requirement to tackle multiple degradations using the same model can lead to high-complexity designs with fixed configuration that lack the adaptability to more efficient alternatives. We propose DyNet, a dynamic family of networks designed in an encoder-decoder style for all-in-one image restoration tasks. Our DyNet can seamlessly switch between its bulkier and lightweight variants, thereby offering flexibility for efficient model deployment with a single round of training. This seamless switching is enabled by our weights-sharing mechanism, forming the core of our architecture and facilitating the reuse of initialized module weights. Further, to establish robust weights initialization, we introduce a dynamic pre-training strategy that trains variants of the proposed DyNet concurrently, thereby achieving a 50% reduction in GPU hours. Our dynamic pre-training strategy eliminates the need for maintaining separate checkpoints for each variant, as all models share a common set of checkpoints, varying only in model depth. This efficient strategy significantly reduces storage overhead and enhances adaptability. To tackle the unavailability of large-scale dataset required in pre-training, we curate a high-quality, high-resolution image dataset named Million-IRD, having 2M image samples. We validate our DyNet for image denoising, deraining, and dehazing in all-in-one setting, achieving state-of-the-art results with 31.34\% reduction in GFlops and a 56.75\% reduction in parameters compared to baseline models. The source codes and trained models are available at https://github.com/akshaydudhane16/DyNet.