CATSplat: Context-Aware Transformer with Spatial Guidance for Generalizable 3D Gaussian Splatting from A Single-View Image
CATSplat is a new transformer-based method for reconstructing 3D scenes from one image using Gaussian Splatting. It solves monocular reconstruction limits by combining textual guidance from language models with spatial guidance derived from depth maps. This dual approach enhances feature learning, leading to state-of-the-art single-view 3D scene reconstruction performance.