AU-IQA
收藏资源简介:
AU-IQA是一个专门用于评估人工智能增强用户生成内容(AI-UGC)感知质量的基准数据集。该数据集由4800张由三种代表性增强类型(超分辨率、低光增强和去噪)生成的AI-UGC图像组成。数据集的建设过程包括选择高质量的UGC图像,对其进行合成降质处理,然后使用不同的AI增强模型进行修复。每个图像都由五名训练有素的标注者使用平均意见得分(MOS)进行标注,以评估其感知质量。AU-IQA旨在解决现有感知质量评估模型在AI-UGC场景下的泛化能力和可靠性问题,促进AI-UGC感知质量评估方法的发展。
AU-IQA is a benchmark dataset specifically designed for evaluating the perceptual quality of Artificial Intelligence-enhanced User-Generated Content (AI-UGC). This dataset consists of 4800 AI-UGC images generated from three representative enhancement types: super-resolution, low-light enhancement, and denoising. The construction process of the dataset includes selecting high-quality UGC images, performing synthetic degradation processing on them, and then restoring them using different AI enhancement models. Each image is annotated by five well-trained annotators using Mean Opinion Score (MOS) to evaluate its perceptual quality. AU-IQA aims to address the generalization ability and reliability issues of existing perceptual quality assessment models in the AI-UGC scenario, and promote the development of AI-UGC perceptual quality assessment methods.

- 1AU-IQA: A Benchmark Dataset for Perceptual Quality Assessment of AI-Enhanced User-Generated Content上海交通大学(Shanghai Jiao Tong University) · 2025年



