Julian Wyatt

Julian Wyatt

Doctoral Student

Student, St Anne's College

E: julian.wyatt@cs.ox.ac.uk

I: Google Scholar

See Also:

• Personal Page

Interests

My research interests are:

• Self-supervised methods
• Anomaly detection
• Generative Models
• Multimodal deep learning

Biography

Julian is a DPhil student in Computer Science, working in the OxMedIS research group. He previously completed an integrated Masters in Computer Science at Durham, which led to his persuit of a DPhil.

His research focuses on self-supervised machine learning methods and applications within the medical domain. Mainly applied to medical landmark detection: a method used by clinicians to produce explainable diagnosis. The below is an example of the Gaussian heatmaps used for landmark detection.

Julian's first year culminated in winning the Cephalometric Landmark Detection Challenge @ MICCAI 2024. His approach utilised an RCNN and novel augmentation techniques to align cephalograms from diverse domains. See below a partial figure from the paper with yellow lines between prediction (error <= 2mm blue points and error > 2mm red points) and ground truth (green points).

Prior to Julian's Oxford research he worked with Dr Chris Willcocks and published work on unsupervised anomaly detection with diffusion models at CVPR workshops 2022 (AnoDDPM). The gifs highlight that when using simplex noise, the anomaly can be removed in fewer steps than Gaussian noise. Therefore, the underlying image structure is less corrupted, leaving more of the original scan.

For further information about Julian, visit his Personal Page: www.julianwyatt.co.uk

Selected Publications

View AllManage publications

• A Handful of Data: Evaluating Few–Shot Incremental Landmark Detection

  Nikil Patel‚ Allison Clement‚ Julian Wyatt‚ Roberto Di Via‚ Davide Marinelli‚ Massimiliano Ciranni‚ Vito Paolo Pastore and Irina Voiculescu

  In Proceedings of the International Conference on Image Analysis and Processing (ICIAP). Springer. 2025.

  Details about A Handful of Data: Evaluating Few–Shot Incremental Landmark Detection | BibTeX data for A Handful of Data: Evaluating Few–Shot Incremental Landmark Detection | Link to A Handful of Data: Evaluating Few–Shot Incremental Landmark Detection

• Optimising for the Unknown: Domain Alignment for Cephalometric Landmark Detection

  Julian Wyatt and Irina Voiculescu

  Winner of the Landmark Detection Challenge at the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI). October, 2024.

  Details about Optimising for the Unknown: Domain Alignment for Cephalometric Landmark Detection | BibTeX data for Optimising for the Unknown: Domain Alignment for Cephalometric Landmark Detection | Link to Optimising for the Unknown: Domain Alignment for Cephalometric Landmark Detection

• Anoddpm: Anomaly detection with denoising diffusion probabilistic models using simplex noise

  Julian Wyatt‚ Adam Leach‚ Sebastian M Schmon and Chris G Willcocks

  Pages 650–656. 2022.

  Details about Anoddpm: Anomaly detection with denoising diffusion probabilistic models using simplex noise | BibTeX data for Anoddpm: Anomaly detection with denoising diffusion probabilistic models using simplex noise | Link to Anoddpm: Anomaly detection with denoising diffusion probabilistic models using simplex noise

Supervisor