AIGCOIQA2024: Perceptual Quality Assessment of AI Generated Omnidirectional Images

TL;DR

This work addresses the absence of IQA criteria for AI-generated omnidirectional images by introducing AIGCOIQA2024, a database of 300 ERP omnidirectional images generated from 5 AIGC models using 25 prompts. It captures human preferences across three dimensions—quality, comfortability, and correspondence—via a subjective study guided by ITU-R BT.500-14, and benchmarks 19 NR-IQA models on these multi-dimensional ratings using SRCC, PLCC, and KRCC. The study reveals that existing handcrafted metrics perform poorly and even state-of-the-art deep models struggle to predict all three dimensions simultaneously, highlighting the need for multi-faceted IQA approaches that incorporate authenticity, comfort, and text-image alignment. By releasing AIGCOIQA2024, the authors provide a valuable resource for developing specialized omnidirectional AIGC IQA methods tailored to VR/AR contexts and promote further research into perceptual quality assessment of AI-generated immersive content.

Abstract

In recent years, the rapid advancement of Artificial Intelligence Generated Content (AIGC) has attracted widespread attention. Among the AIGC, AI generated omnidirectional images hold significant potential for Virtual Reality (VR) and Augmented Reality (AR) applications, hence omnidirectional AIGC techniques have also been widely studied. AI-generated omnidirectional images exhibit unique distortions compared to natural omnidirectional images, however, there is no dedicated Image Quality Assessment (IQA) criteria for assessing them. This study addresses this gap by establishing a large-scale AI generated omnidirectional image IQA database named AIGCOIQA2024 and constructing a comprehensive benchmark. We first generate 300 omnidirectional images based on 5 AIGC models utilizing 25 text prompts. A subjective IQA experiment is conducted subsequently to assess human visual preferences from three perspectives including quality, comfortability, and correspondence. Finally, we conduct a benchmark experiment to evaluate the performance of state-of-the-art IQA models on our database. The database will be released to facilitate future research.

Figures (6)

• Figure 1: An overview of the AIGCOIQA2024 database.
• Figure 2: An example of the subjective experiment interface in unity, participants can use the cursor to click on the score box to select the "quality" score.
• Figure 3: (a) MOS distribution of quality score.(b) MOS distribution of comfortability score.(c) MOS distribution of correspondence score.
• Figure 4: Kernel distribution of four selected features of three databases: AIGCOIQA2024, SUN360 sun360, Matterport3D matterport.
• Figure 5: Relative range and Coverage uniformity of the four selected features on the three databases.