Postdoc in Computer Vision for AI-based Segmentation and Quantification 3D micro-CT data for Natural Heritage

Technical University of Denmark; Kongens Lyngby; On-site; DKK 380,000 - 520,000; Full time

Job summary

The Technical University of Denmark invites applications for a postdoc to advance segmentation and analysis methods for a robotic micro-CT system that digitizes natural heritage collections.

You will work on computer vision for 3D volumetric scans, contribute to open-source software, and collaborate with leading researchers in physics and natural history while publishing your results and advancing scalable, automated workflows.

Qualifications

• Early-career researcher in advanced image segmentation methods.

Responsibilities

• Develop new segmentation methods for large-scale analysis of micro-CT data.
• Develop and validate methods for efficient human-in-the-loop segmentation of CT images.
• Explore the use of self-supervised foundation models for segmentation with limited labels.
• Establish reference benchmark datasets for natural heritage.
• Collaborate with domain experts in natural heritage and evolutionary biology.
• Contribute to and be part of open-source scientific software communities around the Core Imaging Library and qim3D packages.
• Publish scientific articles within computer vision, data, and applications in collaboration with colleagues in physics and natural history.

Skills

Image segmentation

Computer vision

3D imaging

Tools

Core Imaging Library

qim3D

Job description

Are you excited about scientific AI and developing efficient computer vision methods for analysis of complex 3D scans of natural heritage samples? Join our interdisciplinary team as a postdoc to develop advanced segmentation and other analysis methods for a novel robotic micro-CT system that will transform how natural history collections are digitized and studied. You will have the opportunity to conduct high-impact research in computer vision for 3D volumetric images, contribute to widely used open-source software, collaborate with leading researchers in physics and natural history, and publish your work while helping build technology with lasting scientific impact.

Responsibilities and qualifications

We are looking for a highly motivated early-career researcher in advanced image segmentation methods to join our imaging research group and project "Natural Heritage 3D". This collaborative project between Visual Computing @ DTU Compute, DTU Physics, Copenhagen University, and the Natural History Museum aims to develop a fully automated robotic computed tomography system along with computational methods to enable 3D digitization of large collections of natural history specimens.

The Natural History Museum has millions of natural specimens collected over 400 years, which provide an enormous source of information on natural history. CT scans provide a unique method to visualize internal features in 3D in a fully non-destructive manner. In an effort towards digitizing collections and unlocking hidden information on an unprecedented scale, an automated CT system is being developed to scan and analyse large quantities of such specimens. Selected science cases include seal and polar bear craniums as well as snakes and lizards preserved in alcohol.

In addition to experimental CT scan equipment, a key component is the computational pipeline to turn acquired CT projections through reconstruction and image analysis into digital volumes and extract relevant quantitative data. Given the large quantities of specimens, scanning needs to be fast and automated, which demands state-of-the-art reconstruction and analysis methods to handle the data arising from very fast scans taken with few and noisy projections. The focus of this postdoc position is the segmentation and quantification steps, conducting research, and developing suitable analysis methods to produce measurements of the structures in the 3D volumetric images of natural heritage specimens. Our robotic micro-CT system at the DTU 3D Imaging Center enables scanning of thousands of samples, so we will focus on scalability and statistics at a scale beyond current state-of-the-art.

Your primary activities include:

• Develop new segmentation methods for large-scale analysis of micro-CT data.
• Develop and validate methods for efficient human-in-the-loop segmentation of CT images.
• Explore the use of self-supervised foundation models for segmentation with limited labels.
• Establish reference benchmark datasets for natural heritage.
• Collaborate with domain experts in natural heritage and evolutionary biology.
• Contribute to and be part of open-source scientific software communities around the Core Imaging Library and qim3D packages.
• Publish scientific articles within computer vision, data, and applications in collaboration with colleagues in physics and natural history.