AIGCOIQA2024: Perceptual Quality Assessment of AI-Generated Omnidirectional Images This paper presents AIGCOIQA2024, a large-scale database for assessing the perceptual quality of AI-generated omnidirectional images. The database contains 300 omnidirectional images generated using five AI-generated content (AIGC) models and 25 text prompts. A subjective IQA experiment was conducted to assess human visual preferences from three perspectives: quality, comfortability, and correspondence. A benchmark experiment was also conducted to evaluate the performance of state-of-the-art IQA models on the database. AI-generated omnidirectional images have unique distortions compared to natural omnidirectional images, and there is no dedicated IQA criteria for assessing them. The AIGCOIQA2024 database was established to address this gap and facilitate future research. The database includes 300 omnidirectional images generated using five AIGC models and 25 text prompts. The images were generated using various models, including MVDiffusion, Text2Light, DALLE, omni-inpainting, and a fine-tuned Stable Diffusion model. The subjective experiment involved 20 participants who scored the images on three dimensions: quality, comfortability, and correspondence. The results showed that the database encompasses a broad range of scores, indicating its diversity. The database was also compared with other natural omnidirectional databases, such as Matterport3D and SUN360, to evaluate the distribution and uniformity of the features. The paper also presents a benchmark experiment using 19 state-of-the-art no-reference (NR) IQA models. The results showed that deep learning-based models generally outperform hand-crafted models, but their performance is still not entirely satisfactory. The models generally perform better in the quality dimension but worse in the comfortability and correspondence dimensions. The paper concludes that the assessment of AI-generated omnidirectional images must be performed from multiple dimensions, including quality, comfortability, and correspondence. The AIGCOIQA2024 database provides a comprehensive benchmark for future research in this area.AIGCOIQA2024: Perceptual Quality Assessment of AI-Generated Omnidirectional Images This paper presents AIGCOIQA2024, a large-scale database for assessing the perceptual quality of AI-generated omnidirectional images. The database contains 300 omnidirectional images generated using five AI-generated content (AIGC) models and 25 text prompts. A subjective IQA experiment was conducted to assess human visual preferences from three perspectives: quality, comfortability, and correspondence. A benchmark experiment was also conducted to evaluate the performance of state-of-the-art IQA models on the database. AI-generated omnidirectional images have unique distortions compared to natural omnidirectional images, and there is no dedicated IQA criteria for assessing them. The AIGCOIQA2024 database was established to address this gap and facilitate future research. The database includes 300 omnidirectional images generated using five AIGC models and 25 text prompts. The images were generated using various models, including MVDiffusion, Text2Light, DALLE, omni-inpainting, and a fine-tuned Stable Diffusion model. The subjective experiment involved 20 participants who scored the images on three dimensions: quality, comfortability, and correspondence. The results showed that the database encompasses a broad range of scores, indicating its diversity. The database was also compared with other natural omnidirectional databases, such as Matterport3D and SUN360, to evaluate the distribution and uniformity of the features. The paper also presents a benchmark experiment using 19 state-of-the-art no-reference (NR) IQA models. The results showed that deep learning-based models generally outperform hand-crafted models, but their performance is still not entirely satisfactory. The models generally perform better in the quality dimension but worse in the comfortability and correspondence dimensions. The paper concludes that the assessment of AI-generated omnidirectional images must be performed from multiple dimensions, including quality, comfortability, and correspondence. The AIGCOIQA2024 database provides a comprehensive benchmark for future research in this area.