Researcher - Statistical 3D Model of Quadrupeds

Researcher - Statistical 3D Model of Quadrupeds

Max Planck Institute for Intelligent Systems

Tübingen, Germany

The project

This project is aimed at bridging the gap between newly developed state of the art statistical model(s) for quadrupeds, and the physical word. We are collaborating with the Swedish University of Agricultural Sciences and National Research Council of Italy to develop anatomically accurate 3D model, which can be used for biological applications based on one of the biggest datasets of animal scans.

The position

Your task will be to create a method for extracting mesh representations of shape and pose from image data by leveraging existing methods and/or developing novel approaches, as part of the larger goal of bringing our model(s) into active adoption. Specifically, this includes (but is not limited to):

• Creation of generative methods for hyperrealistic scenes and extraction of high-fidelity novel view;
• Training of multi-view and single-view mesh extraction models;
• Validation of the method in a real-life biological context.

Skills and requirements

• Strong background in computer vision, with particular experience in human and animal shape modeling;
• Experience with 3D vision topics such as 3D mesh models, statistical shape modeling and articulated pose estimation;
• Solid ML/AI foundation;
• Strong programming skills in Python (CUDA is a plus);
• Hands-on experience with 3D and image-based computer vision;
• Experience implementing 3D computer vision algorithms and training end-to-end machine learning models (from data collection and design to evaluation);
• Knowledge and practical experience with deep learning, including frameworks such as PyTorch;
• MSc is required in addition a few years of job experience or PhD;
• Good oral and written communication skills in English.