Advanced 3D Gaussian Splatting with statistical linearization

Research topic/area

Computer science

Type of thesis

Bachelor / Master

Start time

-

Application deadline

31.01.2027

Duration of the thesis

6 months

Description

3D Gaussian Splatting (3DGS) has recently revolutionized the field of novel view synthesis by offering state-of-the-art rendering quality at real time speeds using only 2D images. By representing 3D scenes as a collection of anisotropic 3D Gaussians, it circumvents the high computational costs of traditional neural radiance fields.

However, the current mathematical formulation has limitations. To project these 3D Gaussians onto a 2D image plane, 3DGS explicitly linearizes the nonlinear perspective projection by applying a first-order Taylor series expansion around the Gaussian mean. This local affine approximation introduces projection errors, especially for highly distorted camera models or at the edges of the field of view.

In this thesis, a novel approach will be developed inspired by the field of nonlinear state estimation. Instead of relying on a first-order Taylor expansion, this project will use statistical linearization to approximate the projection. This should enhance reconstruction quality by reducing projection errors and enable highly distorted camera models.

Requirement

Requirements for students

• Students with a background in computer science, mathematics, electrical engineering, mechatronics or other engineering majors. It requires a high level of motivation and the ability to work independently and in a structured manner. Prior knowledge of computer graphics, Python and (CUDA) would be helpful but are not essential.

Faculty departments

• Engineering sciences
  Electrical engineering & information technologies
  Geodesy & geoinformatics
  Informatics
  Mechanical engineering
  Mechatronics & information technologies
  Optics & photonics
  Mobility and Infrastructure
  Mechanical Engineering
  Mobility Systems Engineering and Management
  Remote Sensing and Geoinformatics
  Information System Engineering and Management
  Computer Science
  Electrical Engineering and Information Technology
  Mechatronics and Information Technology
  Medical technology
  Photon Science and Technology
• Natural sciences and Technology
  Mathematics
  Physics
  Mathematics in Technology
  Physics
  Techno-Mathe­matics

Supervision

Title, first name, last name

Jiachen Zhou

Organizational unit

Institut für Anthropomatik und Robotik (IAR) - Intelligent Sensor-Actuator-Systems (ISAS)

Email address

jiachen.zhou@kit.edu

Link to personal homepage/personal page

Website

Application via email

Application documents

• Curriculum vitae
• Grade transcript

E-Mail Address for application
Senden Sie die oben genannten Bewerbungsunterlagen bitte per Mail an jiachen.zhou@kit.edu


Back