Abstract:Recent advancements in the field of No-Reference Image Quality Assessment (NR-IQA) using deep learning techniques demonstrate high performance across multiple open-source datasets. However, such models are typically very large and complex making them not so suitable for real-world deployment, especially on resource- and battery-constrained mobile devices. To address this limitation, we propose a compact, lightweight NR-IQA model that achieves state-of-the-art (SOTA) performance on ECCV AIM UHD-IQA challenge validation and test datasets while being also nearly 5.7 times faster than the fastest SOTA model. Our model features a dual-branch architecture, with each branch separately trained on synthetically and authentically distorted images which enhances the model's generalizability across different distortion types. To improve robustness under diverse real-world visual conditions, we additionally incorporate multiple color spaces during the training process. We also demonstrate the higher accuracy of recently proposed Kolmogorov-Arnold Networks (KANs) for final quality regression as compared to the conventional Multi-Layer Perceptrons (MLPs). Our evaluation considering various open-source datasets highlights the practical, high-accuracy, and robust performance of our proposed lightweight model. Code: <a class="link-external link-https" href="https://github.com/nasimjamshidi/LAR-IQA" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Develop a lightweight, accurate and robust no - reference image quality assessment (NR - IQA) model, making it suitable for real - time applications and resource - constrained mobile devices**. Specifically, although existing NR - IQA models show high performance on multiple open - source datasets, these models are usually very large and complex and are not suitable for practical deployment, especially on mobile devices with limited resources and battery. To address this limitation, the author proposes a compact and lightweight NR - IQA model, which achieves state - of - the - art performance on the validation and test datasets of the ECCV AIM UHD - IQA challenge, while being nearly 5.7 times faster than the fastest existing model. ### Main problems and challenges 1. **Model complexity and computational resources**: Traditional deep - learning NR - IQA models are accurate but usually have high computational complexity and are difficult to achieve real - time assessment on resource - constrained devices. 2. **Generalization ability**: Existing NR - IQA models perform poorly when dealing with different types of distortion, especially under real - world conditions. 3. **Robustness of color space**: Different color spaces have different impacts on image quality, so the model needs to be able to adapt to multiple color spaces. ### Solutions The author proposes a lightweight NR - IQA model (LAR - IQA) with a two - branch architecture. The main innovations include: - **Two - branch architecture**: One branch is trained on synthetically distorted images, and the other branch is trained on real - distorted images to enhance the generalization ability of the model. - **Kolmogorov - Arnold Networks (KAN)**: Used for the final quality regression. Compared with the traditional multi - layer perceptron (MLP), KAN provides higher accuracy and robustness. - **Multi - color - space training**: By introducing multiple color spaces during the training process, the robustness of the model under different visual conditions is improved. ### Experimental results The experimental results show that the performance of this model on multiple public datasets is better than existing methods, and it is more efficient in terms of computational resource consumption, making it suitable for real - time applications and mobile devices. Through these improvements, the LAR - IQA model not only reaches the state - of - the - art level in performance but also has better feasibility and efficiency in practical applications.

LAR-IQA: A Lightweight, Accurate, and Robust No-Reference Image Quality Assessment Model

MSLIQA: Enhancing Learning Representations for Image Quality Assessment through Multi-Scale Learning

ARNIQA: Learning Distortion Manifold for Image Quality Assessment

No-reference image quality assessment based on residual neural networks (ResNets)

MobileIQA: Exploiting Mobile-level Diverse Opinion Network For No-Reference Image Quality Assessment Using Knowledge Distillation

Naturalness-Aware Deep No-Reference Image Quality Assessment

No‐reference image quality assessment via a dual‐branch residual network

No-Reference Image Quality Assessment: an Attention Driven Approach

No-reference Video Quality Assessment Based on Spatiotemporal Slice Images and Deep Convolutional Neural Networks

A Data-Driven No-Reference Image Quality Assessment via Deep Convolutional Neural Networks.

Less is More: Learning Reference Knowledge Using No-Reference Image Quality Assessment

An Accurate Deep Convolutional Neural Networks Model for No-Reference Image Quality Assessment

Making a “Completely Blind” Image Quality Analyzer

You Only Train Once: A Unified Framework for Both Full-Reference and No-Reference Image Quality Assessment

ConIQA: A deep learning method for perceptual image quality assessment with limited data

Hierarchical Curriculum Learning for No-Reference Image Quality Assessment

Nested Error Map Generation Network for No-Reference Image Quality Assessment.

Perceptual quality assessment for no-reference image via optimization-based meta-learning

Deep Neural Networks for No-Reference and Full-Reference Image Quality Assessment

No reference image quality assessment metric via multi-domain structural information and piecewise regression

Perceptual Attacks of No-Reference Image Quality Models with Human-in-the-Loop