
TransUNet for Cross-Domain Semantic Segmentation of Urban Scenery
Applying a transformer–CNN architecture to urban scenery, and studying how simulated images can complement real-world training data.
Published research and technical perspectives on building AI that works outside the demonstration.

Applying a transformer–CNN architecture to urban scenery, and studying how simulated images can complement real-world training data.

An overview of visual and audio liveness detection, with implementation details including convolutional neural networks and Mel spectrograms for audio.

Investigating coloured edge maps as CNN inputs for oil-palm ripeness grading, with a focus on robustness to changes in lighting.
What changes when vision systems meet real operating conditions.
How analytical agents fit into technical and business workflows.
How to investigate whether a bespoke AI problem is tractable.
Our project summaries offer a first look at the work behind these topics.
Explore our projects