English summary
The project develops GF4Shape, combining geometric and learned features through a diffusion approach for point-cloud shape completion. Students will design a geometric-feature-based regularizer, implement and evaluate the method on indoor and urban outdoor environments, and compare it with M1 results. The planned work includes literature review, prototyping, dataset evaluation, report writing, and a final presentation.
Overview
- Project details
- GF4Shape: Geometric Features for Point Cloud Shape Completion
Source excerpt · Page 1
- “Project title: GF4Shape: Geometric Features for Point Cloud Shape Completion” Page 1
- Organization
- Not stated in the PDF
- Project formats
- Not stated in the PDF
- Degree levels
- Not stated in the PDF
- Project goal
- Leverage geometric priors as a structural regularizer for shape completion, bridging the gap between accuracy and generalization in complex environments from aerial to terrestrial acquisition.
Source excerpt · Page 1
- “Our goal is to leverage geometric priors as structural regularizer for shape completion, aiming to bridge the gap between accuracy and generalization in complex environments from aerial to terrestrial acquisition.” Page 1
Topics and work
- Subjects
- 3D point cloud shape completion
Source excerpt · Page 1
- “3D shape completion in point clouds remains fundamentally unsolved challenge.” Page 1
- geometric features and priors
Source excerpt · Page 1
- “Our preliminary research indicates that integrating explicit geometric features (e.g., anisotropy, roughness, planarity) with learned features substantially improves the network’s performance for semantic segmentation accelerating convergence and enhancing robustness across datasets.” Page 1
- learned features
Source excerpt · Page 1
- “Develop a method, GF4Shape, that fuses geometric features and learned features based on the diffusion approach.” Page 1
- diffusion approach
Source excerpt · Page 1
- “Develop a method, GF4Shape, that fuses geometric features and learned features based on the diffusion approach.” Page 1
- 3D point cloud shape completion
- Application areas
- indoor environments
Source excerpt · Page 1
- “Evaluate the GF4Shape shape completion method on indoor and urban outdoor environments.” Page 1
- urban outdoor environments
Source excerpt · Page 1
- “Evaluate the GF4Shape shape completion method on indoor and urban outdoor environments.” Page 1
- aerial and terrestrial acquisition
Source excerpt · Page 1
- “Our goal is to leverage geometric priors as structural regularizer for shape completion, aiming to bridge the gap between accuracy and generalization in complex environments from aerial to terrestrial acquisition.” Page 1
- indoor environments
- Methods and tools
- geometric features
Source excerpt · Page 1
- “Develop a method, GF4Shape, that fuses geometric features and learned features based on the diffusion approach.” Page 1
- learned features
Source excerpt · Page 1
- “Develop a method, GF4Shape, that fuses geometric features and learned features based on the diffusion approach.” Page 1
- diffusion approach
Source excerpt · Page 1
- “Develop a method, GF4Shape, that fuses geometric features and learned features based on the diffusion approach.” Page 1
- geometric-feature-based regularizer
Source excerpt · Page 1
- “Design geometric-feature-based regularizer for shape completion and compare to the M1 results.” Page 1
- fitting datasets
Source excerpt · Page 2
- “Evaluation on fitting datasets” Page 2
- geometric features
- Activities
- Familiarisation with relevant papers and literature review
Source excerpt · Page 1
- “Familiarisation with relevant papers, literature review” Page 1
- Propose several ideas for approaching the problem
Source excerpt · Page 1
- “Proposal of several ideas of how to approach the problem” Page 1
- Prototype ideas
Source excerpt · Page 1
- “Prototyping of ideas” Page 1
- Select the most promising path
Source excerpt · Page 2
- “Select the most promising path” Page 2
- Achieve a basic implementation
Source excerpt · Page 2
- “Achieve a basic implementation” Page 2
- Develop GF4Shape to fuse geometric and learned features
Source excerpt · Page 1
- “Develop a method, GF4Shape, that fuses geometric features and learned features based on the diffusion approach.” Page 1
- Design a geometric-feature-based regularizer and compare it to M1 results
Source excerpt · Page 1
- “Design geometric-feature-based regularizer for shape completion and compare to the M1 results.” Page 1
- Evaluate the shape completion method on indoor and urban outdoor environments
Source excerpt · Page 1
- “Evaluate the GF4Shape shape completion method on indoor and urban outdoor environments.” Page 1
- Finalize the implementation
Source excerpt · Page 2
- “Finalizing of implementation” Page 2
- Evaluate on fitting datasets
Source excerpt · Page 2
- “Evaluation on fitting datasets” Page 2
- Write a report
Source excerpt · Page 2
- “Report writing” Page 2
- Give a final presentation
Source excerpt · Page 2
- “Final presentation” Page 2
- Familiarisation with relevant papers and literature review
- Kinds of work
- Literature research
Source excerpt · Page 1
- “Familiarisation with relevant papers, literature review” Page 1
- Software development
Source excerpt · Page 2
- “Achieve a basic implementation” Page 2
- Data analysis and ML
Source excerpt · Page 1
- “Evaluate the GF4Shape shape completion method on indoor and urban outdoor environments.” Page 1
- Literature research
- Expected outputs
- GF4Shape method
Source excerpt · Page 1
- “Develop a method, GF4Shape, that fuses geometric features and learned features based on the diffusion approach.” Page 1
- basic implementation
Source excerpt · Page 2
- “Achieve a basic implementation” Page 2
- geometric-feature-based regularizer
Source excerpt · Page 1
- “Design geometric-feature-based regularizer for shape completion and compare to the M1 results.” Page 1
- report
Source excerpt · Page 2
- “Report writing” Page 2
- final presentation
Source excerpt · Page 2
- “Final presentation” Page 2
- GF4Shape method
Requirements
- Required skills
- Not stated in the PDF
- Recommended skills
- Not stated in the PDF
- Eligible study fields
- Not stated in the PDF
- Programming in the project
- Some
Source excerpt · Page 2
- “Achieve a basic implementation” Page 2
- Programming prerequisite
- Not stated in the PDF
Practical details
- Team size
- a selection of students from different TUM departments
Source excerpt · Page 2
- “The group of students for this project is a selection of students from different TUM departments” Page 2
- Duration
- 13.04.2026 - 31.08.2026
- Start
- 13.04.2026Normalized date: 2026-04-13
Source excerpt · Page 1
- “13.04.2026 - 14.05.2026” Page 1
- Application deadline
- Not stated in the PDF
- Work location
- Not stated in the PDF
- Location mode
- Not stated in the PDF
- Working language
- data analysis and mathematics
Source excerpt · Page 2
- “the common language to all students is data analysis and mathematics.” Page 2
- data analysis and mathematics
Learning and support
- Learning opportunities
- Students not directly related to data science will benefit by attending Fundamentals of Mathematical Statistics to approach interdisciplinary work
Source excerpt · Page 2
- “Students not directly related to data science will benefit by attending the course Fundamentals of Mathematical Statistics in order to approach interdisciplinary work.” Page 2
- Students not directly related to data science will benefit by attending Fundamentals of Mathematical Statistics to approach interdisciplinary work
- Support offered
- Not stated in the PDF
Application and contacts
- Contacts
- Prof. Dr. Drton
Source excerpt · Page 1
- “Project leads: Prof. Dr. Drton (TUM Department of Mathematics)” Page 1
- Dr. Ricardo Cabra
Source excerpt · Page 1
- “Dr. Ricardo Cabra (Munich Data Science Institute MDSI)” Page 1
- Dr. Yan Xia
Source excerpt · Page 1
- “Scientific lead: Dr. Yan Xia and Prof. Dr. Olaf Wysocki” Page 1
- Prof. Dr. Olaf Wysocki
Source excerpt · Page 1
- “Scientific lead: Dr. Yan Xia and Prof. Dr. Olaf Wysocki” Page 1
- Prof. Dr. Drton
- How to apply
- Not stated in the PDF
- Further information
- Not stated in the PDF