Images, data and models

Every image on this site is either my own work (charts, diagrams, drawings, screenshots of my own software) or comes from a dataset under a free licence. Photos from non-commercial, research-only or unlicensed datasets are never shown, even when a project was evaluated on them.

Dataset images shown

• Vehicle Registration Plates (Roboflow / Augmented Startups): CC BY 4.0
• UK Car Plate VRN dataset (synthetic plates): MIT
• EuroSAT RGB (Helber et al. 2019): MIT
• Sign Language MNIST: CC0 1.0
• The football replay's own canvas and 3D rendering of the IDSSE Bundesliga tracking data (Bassek et al. 2025, DFL / Sportec Solutions): Data CC BY 4.0, credited in the caption; the drawing is own work (MIT)
• Six illusion photos from Wikimedia Commons (credits on /credits/): CC BY 2.0 / CC BY-SA 4.0 / CC0; the figure is shared under CC BY-SA 4.0

Illusion photos (project 14)

From Wikimedia Commons; the figure that combines them with depth predictions is shared under CC BY-SA 4.0.

• Ames room forced perspective by mosso, CC BY 2.0
• 3D street painting by Limnesh Augustine, CC BY-SA 4.0
• Den Helder - Julianaplein, 3D street art 2020 by Lars van der Heide, CC BY-SA 4.0
• Colonnade, trompe l'oeil mural by John Pugh by John Pugh, CC BY-SA 4.0
• Film crew in trompe l'oeil mural, Cinéma Vox, Chamonix by DimiTalen, CC0
• Man photographing himself in a hotel bathroom with mirrors by Pittigrilli, CC BY-SA 4.0

Data used in charts

• Expected-goals chart (project 20): data from StatsBomb open data, under the StatsBomb Public Data User Agreement.
• Football replay (project 20): Bundesliga tracking and event data, Bassek et al. (2025), DFL / Sportec Solutions, IDSSE dataset, CC BY 4.0 (subsampled to 12.5 fps, rounded to the centimetre, events synchronised to the tracking).
• Other charts show my own measurements; each case study names its benchmark and links to the dataset's licence in the code repository.

Software in the live demos

• Virtual painter: MediaPipe Tasks Vision and its hand landmark model, Apache-2.0 (notice in /mediapipe/NOTICE.txt).
• Hard-hat check: ONNX Runtime Web, MIT; the model is my YOLO11n (Ultralytics, AGPL-3.0) trained on Hard Hat Workers (CC BY 4.0).

Full dataset and model licence table: DATA_LICENSES.md.