Speaker: Han Liang
This master's thesis investigates the application of variational methods for reconstructing smooth textured surfaces from point cloud rendered images. By utilizing regularization techniques and incorporating normal map information, the proposed approach aims to generate high-quality surface reconstructions that preserve the scene's geometry and structure. The method employs depth maps and corresponding normal maps as input data. A custom loss function is designed to balance smoothness, edge preservation, and normal map consistency when smoothing the depth map. The smoothed depth map then serves as a guide to refine the textured surface, promoting coherent texture information among neighboring pixels with close depth values. The objective of this work is to assess the performance of the proposed technique in reconstructing smooth textured surfaces from point cloud rendered images while preserving important geometric features. In future work, the approach will be extended to smooth path traced images.