My complete master thesis "Patch-Based Inpainting of 3D Gaussian Splatting" is available in the TU Delft repository.

I wrote my master thesis at the Computer Graphics & Visualisation group at TU Delft, where I did research on inpainting 3D Gaussian Splatting (3DGS) scenes. The original 3DGS paper had been published just months before my research started and the key idea behind my work was that we could use the explicit nature of the 3D Gaussians (compared to other methods like NeRF) to perform inpainting based on the content of the scene, by copying similar parts from elsewhere in the scene and blending the copied parts together. My method was based on the PatchMatch image inpainting algorithm.

It was exciting to work with a method as novel 3DGS and through my research I gained a deeper understanding about these methods to produce and edit photorealistic 3D representations in general. Besides that I worked a lot with PyTorch to implement my ideas, which improved my skills regarding vectorisation of calculations and optimisation of performance-sensitive code.