The paper introduces the 3DGCQA dataset, a novel quality assessment database for 3D AI-generated contents (3DGCs). The dataset was constructed using 7 representative Text-to-3D generation methods, with 50 fixed prompts generating a total of 313 textured meshes.
The visualization of the generated 3DGCs reveals the presence of 6 common distortion categories, including multifaceted repetition, depth error, roughness, misalignment, geometry loss, and geometry redundancy. To further explore the quality of the 3DGCs, subjective quality assessment was conducted by 40 evaluators, whose ratings showed significant variation in quality across different generation methods.
Additionally, the paper evaluates several existing objective quality assessment algorithms on the 3DGCQA dataset. The results expose limitations in the performance of these algorithms and underscore the need for developing more specialized quality assessment methods tailored to 3DGCs. The 3DGCQA dataset has been open-sourced to provide a valuable resource for future research and development in 3D content generation and quality assessment.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Yingjie Zhou... at arxiv.org 09-12-2024
https://arxiv.org/pdf/2409.07236.pdfDeeper Inquiries