Projects

Rethinking frequency-domain augmentation to close the robustness gap left by standard visual augmentations.

machine-learning computer-vision data-augmentation robustness

A unified, benchmarked codebase for training and comparing well-known self-supervised vision methods (DINO, iBOT, LeJEPA) under identical conditions.

machine-learning self-supervised-learning computer-vision open-source

Combining low-dimensional image manifold models with optimal transport priors to better reconstruct time-evolving images from scarce data.

machine-learning medical-imaging optimal-transport dynamic-imaging

A self-supervised regularisation technique that explicitly aligns representations across views to build stronger 3D medical foundation models.

machine-learning medical-imaging self-supervised-learning segmentation

Computer vision app that measures and compares sail shape from photos, developed with North Sails and Alive Industries.

computer-vision machine-learning mobile-app

A latent-space regularisation framework for U-Net models that improves domain generalisation and continual learning in medical segmentation.

machine-learning medical-imaging segmentation continual-learning

Evaluating MixUp and Auxiliary Fourier Augmentation for out-of-distribution generalisation in cardiac and prostate MRI segmentation.

machine-learning medical-imaging data-augmentation segmentation

Large-scale challenge findings on building brain MRI foundation models that hold up on noisy, real-world clinical data.

machine-learning medical-imaging foundation-models self-supervised-learning

P. Vaish (Mei)