About us

From research.
For the real world.

We are civil engineers, AI researchers, and software developers from TU Munich, the University of the Bundeswehr Munich, and the University of Oxford. Our models are not created in a vacuum. They are based on 9,920 real inspection images, published as dacl10k, the world's largest dataset for automated bridge inspection worldwide.
We build software by engineers for engineers.

Scientifically sound

Our scientific work.

dacl10k

9,920 real inspection images. 13 damage classes. 6 component groups. The benchmark for AI-assisted bridge inspection.

Publication

synth-dacl

Synthetic data for damage detection. Synthetic training data that makes manual labeling of cracks unnecessary. Better results than with real data alone.

Publication

Number Token Loss (NTL)

Number Token Loss

A new training method that teaches language models to understand numbers as numbers, not as words.

Publication

dacl10k

We have released dacl10k, the largest and most diverse dataset for automated bridge inspection, comprising 9,920 real inspection images with detailed annotations. Together with strong baseline models, this benchmark has set a new standard for advancing AI-assisted bridge inspection.

Publication