Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Human Clothing Reconstruction from 3D Scans
To achieve photorealism in digital human representations, the faithful reconstruction of human clothing is a must. The first step towards this goal is the ability to extract garments from high-quality 3D Scans. Modern advances in learned cloth simulation can provide good prior knowledge of the physical behavior of fabric and be used to register garment geometry from scans.
In this project, you are going to work with a state-of-the-art deep learning approach for modeling garment dynamics and apply it to the task of garment registration from scans.
That is, given a 4D scan sequence and a garment geometry, the goal is to register the garment geometry to all frames in the sequence.
Starting as a semester project, your work we’ll be a part of the ongoing CVPR2024 project. Then, it can be extended to a master thesis and possibly another publication.
**Benefits:**
- You will work with a state-of-the-art deep-learning method for garment simulation
- You will work with 4D scan data collected by a modern capturing studio in our lab
- You’ll potentially participate in writing a paper for a major computer vision conference
**Requirements:**
Interested students should have experience with training neural networks and be able to work with modern deep learning frameworks (namely, PyTorch). Background in computer graphics can be a major advantage, but not a necessity.
**Related Work:**
- HOOD: Hierarchical Graphs for Generalized Modelling of Clothing Dynamics https://arxiv.org/abs/2212.07242
- Deep Physics-aware Inference of Cloth Deformation for Monocular Human Performance Capture https://arxiv.org/abs/2011.12866
In this project, you are going to work with a state-of-the-art deep learning approach for modeling garment dynamics and apply it to the task of garment registration from scans.
That is, given a 4D scan sequence and a garment geometry, the goal is to register the garment geometry to all frames in the sequence.
Starting as a semester project, your work we’ll be a part of the ongoing CVPR2024 project. Then, it can be extended to a master thesis and possibly another publication.
**Benefits:**
- You will work with a state-of-the-art deep-learning method for garment simulation - You will work with 4D scan data collected by a modern capturing studio in our lab - You’ll potentially participate in writing a paper for a major computer vision conference
**Requirements:**
Interested students should have experience with training neural networks and be able to work with modern deep learning frameworks (namely, PyTorch). Background in computer graphics can be a major advantage, but not a necessity.
**Related Work:**
- HOOD: Hierarchical Graphs for Generalized Modelling of Clothing Dynamics https://arxiv.org/abs/2212.07242 - Deep Physics-aware Inference of Cloth Deformation for Monocular Human Performance Capture https://arxiv.org/abs/2011.12866
1st part of the project (semester project): devise a baseline method for registering garment geometry to 4D scan data. This part is mainly technical, you’ll work closely with a research assistant working toward a CVPR submission
2nd part of the project (master thesis): improve the baseline method for fine-grained registration of garment geometry to 4D scans and 4D implicit representation sequences extracted from monocular videos.
This part is research-oriented. If good results are achieved it may result in your first-author publication at a major computer vision conference.
1st part of the project (semester project): devise a baseline method for registering garment geometry to 4D scan data. This part is mainly technical, you’ll work closely with a research assistant working toward a CVPR submission
2nd part of the project (master thesis): improve the baseline method for fine-grained registration of garment geometry to 4D scans and 4D implicit representation sequences extracted from monocular videos. This part is research-oriented. If good results are achieved it may result in your first-author publication at a major computer vision conference.
This project will be supervised by Artur Grigorev: https://ait.ethz.ch/people/agrigorev
E-mail: agrigorev@ethz.ch
This project will be supervised by Artur Grigorev: https://ait.ethz.ch/people/agrigorev