T$^3$Bench: Benchmarking Current Progress in Text-to-3D Generation

arXiv:2310.02977v2 Announce Type: replace
Abstract: Recent methods in text-to-3D leverage powerful pretrained diffusion models to optimize NeRF. Notably, these methods are able to produce high-quality 3D scenes without training on 3D data. Due to the open-ended nature of the task, most studies evaluate their results with subjective case studies and user experiments, thereby presenting a challenge in quantitatively addressing the question: How has current progress in Text-to-3D gone so far? In this paper, we introduce T$^3$Bench, the first comprehensive text-to-3D …

arxiv benchmarking cs.cl cs.cv cs.lg current progress text type