Computer Vision
Object detection and tracking, video understanding, vision-language models, document intelligence and OCR, real-time inference on edge devices.
Why we work in this area
In vision models the gap between laboratory results and field results is larger than in any other area. Lighting changes, camera angles drift, line speed increases, hardware is limited. A model performing well on a test set does not mean it will run for hours on a production line without failing.
That is why the weight of our work sits on real-time behaviour and robustness: keeping a model's accuracy while running it on constrained hardware, uninterrupted and with predictable latency.
What we work on
- Object detection, classification, segmentation and multi-object tracking
- Video understanding and temporally consistent analysis
- Vision-language models (VLM), document intelligence and OCR
- Edge deployment: distillation, quantisation, TensorRT acceleration, and measuring the latency–accuracy trade-off
- Data capture through non-standard imaging methods (alternative light sources, spectral imaging)
From our research projects
Our funded projects in this area concentrate on four problems: anomaly detection and visual quality inspection on high-speed production lines, verifying product authenticity in the field with standard mobile devices, detecting counterfeit goods through specialised imaging techniques, and identifying counterfeits on online platforms from their digital traces.
The first three share the same difficulty: producing reliable results under uncontrolled conditions on limited hardware. The last is a different class of problem — deciding without any access to the physical product, by evaluating visual and textual signals together.
A technical assessment for your AI project
Your project's feasibility, risks and timeline are assessed in a technical consultation.