Pruning Overparameterized Multi-Task Networks for Degraded Web Image Restoration
Thomas Katraouras, Dimitrios Rafailidis
2025-10-22
Summary
This paper focuses on improving how images look online, specifically by fixing problems caused when social media platforms compress them. It introduces a new method to make image-fixing programs smaller and faster without losing quality.
What's the problem?
When you upload a picture to social media, the site often shrinks the file size to save space and bandwidth. This compression can make the image look blurry, hazy, or noisy – basically, it degrades the image quality. Existing programs that try to *restore* these damaged images are often very large and require a lot of computing power, making them inefficient.
What's the solution?
The researchers developed a technique called MIR-L that dramatically reduces the size of these image restoration programs. They do this by identifying and keeping only the most important parts of the program – the parts that contribute the most to fixing the image. It’s like trimming away unnecessary branches from a tree to help the main trunk grow stronger. This 'trimming' happens in multiple steps, carefully removing less important connections within the program while making sure the remaining parts still work well. They tested it on fixing images affected by rain, haze, and noise.
Why it matters?
This work is important because it allows for high-quality image restoration to be done on devices with limited processing power, like phones or older computers. By making these programs smaller and faster, it improves the user experience online by ensuring images look their best, even after being compressed by social media platforms. It also makes it more practical to use these restoration techniques in a wider range of applications.
Abstract
Image quality is a critical factor in delivering visually appealing content on web platforms. However, images often suffer from degradation due to lossy operations applied by online social networks (OSNs), negatively affecting user experience. Image restoration is the process of recovering a clean high-quality image from a given degraded input. Recently, multi-task (all-in-one) image restoration models have gained significant attention, due to their ability to simultaneously handle different types of image degradations. However, these models often come with an excessively high number of trainable parameters, making them computationally inefficient. In this paper, we propose a strategy for compressing multi-task image restoration models. We aim to discover highly sparse subnetworks within overparameterized deep models that can match or even surpass the performance of their dense counterparts. The proposed model, namely MIR-L, utilizes an iterative pruning strategy that removes low-magnitude weights across multiple rounds, while resetting the remaining weights to their original initialization. This iterative process is important for the multi-task image restoration model's optimization, effectively uncovering "winning tickets" that maintain or exceed state-of-the-art performance at high sparsity levels. Experimental evaluation on benchmark datasets for the deraining, dehazing, and denoising tasks shows that MIR-L retains only 10% of the trainable parameters while maintaining high image restoration performance. Our code, datasets and pre-trained models are made publicly available at https://github.com/Thomkat/MIR-L.