AI-driven visual analytics, automated quality inspection, and defect & surface segmentation for research and industry.
Few-shot deployment reaches a new line on as few as 10 annotated images, where a conventional pipeline needs thousands. Method and evaluation in our publications.
Grounded in
01 Capabilities
Six research lines, all pointed at the same problem: turning raw industrial imagery into a decision an engineer can act on.
Flagship
Defect detection, anomaly spotting, and real-time quality assurance pipelines for high-throughput manufacturing — built to hold accuracy at line speed, to say clearly when they are uncertain, and to reach a new line on minimal annotated data.
Survey and sensor data turned into maintained digital models of real structures — and of how their condition changes over time.
High-precision 3D reconstruction, point-cloud analysis, and metrology for complex structural environments.
Single-pixel imaging, compressive sensing, and learned image reconstruction — recovering a usable image before anything has to interpret it.
Attention-refined feature pyramids, adversarial and probabilistic models — for defects that are thin, faint, or genuinely ambiguous.
Damage detection and condition assessment across composites, concrete, and ageing infrastructure.
02 Application areas
The same perception stack, tuned to the failure modes and tolerances each sector actually cares about.
03 Newsroom
We welcome Wissem Chachoua and Anaïs Bellagha to the lab for an eight-week research internship.
A GMM-based VAE with normalising flow for effective stochastic segmentation, presented at NeurIPS 2025.
Our deep context learning model with anomalous trend alarming, published in Results in Engineering.
A feature pyramid network with a self-guided attention refinement module for crack segmentation.
Inspection is where most people find us, but the work is broader: reconstructing a surface, recovering an image the sensor barely caught, tracking how an asset changes. If it can be imaged, it can be measured — and we take on the ones without an off-the-shelf answer.