The problem of 3D Reconstruction from images is undergoing a fundamental transfor-
mation, driven by continuous advances in Implicit Neural Representations (such as Neural
Radiance Fields) and 3D Gaussian splatting. In this setting, we represent a scene as a volume
and render using volume rendering. However, directly extracting detailed surfaces from such
fuzzy density representation poses challenges due to insufficient surface regularization. This
problem led researchers to experiment with new representations that enable these regulariza-
tions, such as 2D Gaussians, which have shown promising results for surface extraction. This
work is a study on both INRs and 2D Gaussian Splatting. Firstly, we will give a detailed
analysis of the theoretical background of Gaussian Splatting. Secondly, we will also show that
mixing 2DGS with depth can significantly improve geometric accuracy by the order of 10%
compared with no depth information. Next, we will show an application for extracting an INR
from 2D Gaussians. Finally, we will conclude this dissertation by focusing on 360° rotations
around an object of interest. We will present a method to segment between the background
and foreground Gaussians and, with this segmentation, extract a 360° panorama.