3D Gaussian Inpainting with Depth-Guided Cross-View Consistency

Figure 1. Given multi-view images of a scene and the object masks describing the object to be removed, our 3D Gaussian Inpainting with Depth-Guided Cross-View Consistency produces high-fidelity cross-view inpainting results. Compared with current state-of-the-arts such as SPIn-NeRF [24], Gaussian Grouping [41], and GScream [35], inpainting results of our method not only preserve visible background contents but also exhibit satisfactory consistency across camera views.