3D indoor scene assessment via layout plausibility

Xinyan Yang, Fei Hu, Shaofei Liu, Long Ye, Ye Wang, Guanghua Zhu, Jiyin Li

Abstract: Highlights•3D-SPAD-MI dataset is established. It adds 6 new scene categories of 900 sample data so that the dataset can cover more tasks about 3D scene plausibility assessment tasks.•In 3D scene plausibility assessment task, a novel semantic and visual feature fusion network 3D-SPAN-M is proposed. It adopts the cross-modal fusion multi-task learning method to improve the 3D scene assessment performance significantly.•For assessment in small-scaled data, 3D-SPAN-F is proposed. It assesses new categories of 3D scene plausibility by a few-shot learning network.