Fig. 6
From: Single view generalizable 3D reconstruction based on 3D Gaussian splatting
Cross-Dataset Generalization Experiment. Although the images in the KITTI and NYU test sets have a larger baseline comparison, our network demonstrates strong reconstruction capabilities for target viewpoints and performs well in synthesizing new perspectives. We conducted experiments on three distinct scenes from the NYU and KITTI datasets, respectively. The experimental results indicate that SVG3D outperforms other network models.
