plAIdata: AI based PLanning and monitoring of buildings towards Data sovereignty
RUNNING TIME:
01/2025
—
12/2026
Total project duration:
2 Years
Photo created with generative AI, photo: JOANNEUM RESEARCH
The project
The application of AI has large potential of facilitating the documenting and monitoring of buildings and other human-made structures. Collecting and labelling the data needed for training these AI methods is costly, and existing datasets are often in compliance with the EU legislation, and make companies dependent on the providers of these datasets. plAIdata researches methods that enable Austrian companies to make their own image data usable for AI, and generate synthetic data for training.
Research groups
Downloads
Customers
Project partners
DIBIT Messtechnik GmbH
Universität Wien - Institut für Innovation und Digitalisierung im Recht
xCAD Solutions GmbH
Project details
The Austrian companies DIBIT Messtechnik GmbH and xCAD Solutions GmbH bring their know how and technology framework for the respective use cases (tunnel construction and interior planning). The Department of Innovation and Digitalisation in Law of the University of Vienna analyses questions of AI regulation, data protection and copyright. JOANNEUM RESEARCH coordinates the project and researches AI methods for both use cases.
The availability of datasets under conditions that make their use legally and commercially feasible is a prerequisite to leverage the benefits of AI-based methods in the target use cases – interior planning and tunnel construction – for European companies, and to make them independent of actors outside the EU.
plAIdata will address two strategies:
- Develop tools for the generation of synthetic data, such as based on rendering 3D scenes, for which both the images and the annotations of the objects shown in the scenes can be generated automatically. While the use of synthetic data allows generating large datasets at acceptable costs, it needs to be complemented by appropriate methods for transferring the learned models to real data, bridging the domain gap. The use of data- efficient 2D, 3D and hybrid methods to generate synthetic data is not only a requirement to keep annotation costs low, but will also enable for more energy-efficient training methods.
- Research approaches for semi-automatic and data-efficient annotation of real data. By leveraging advances in vision-language foundation models and methods capable of learning from small sets of annotated samples, plAIdata will reduce the costs for labelling training data and customizing models for visual detection and segmentation.
plAIdata also analyses the legal framework around creation and usage of such datasets, in order to enable Austrian companies to use them in conformity with European legislation. The project will develop guidelines that allow a structured approach to data collection and generation, thus providing legal clarity.
