之后,SDF 3D transformer以从粗到细的方式聚合这三个特性。在粗level和中等level上,3D transformer的输出为两个占用体素\mathbf{O}^{2}, \mathbf{O}^{1},而在精细level上,输出是预测的TSDF体素\mathbf{S}^{0}。粗占用体素\mathbf{O}^{2}和中等占用体素\mathbf{O}^{1}存储体素的占用值o \in[0,1],...
尤其是,在ScanNet数据集上,重建精度提高了41.8%,重建完整度提高了25.3%。 3D-Former: Monocular Scene Reconstruction with SDF 3D Transformers Weihao Yuan, Xiaodong Gu, Heng Li, Zilong Dong, Siyu Zhu 项目主页:Monocular Scene Reconstruction with 3D SDF Transformers 论文链接:3D Former: Monocular Scene Reco...