Research
Our work centres on self-supervised and multimodal representation learning, and on taking those models to concrete problems with partners in health, industry and cultural heritage.
Research lines
-

Medical Imaging
AI is advancing fast in language and vision, and healthcare stands to benefit the most. We build medical imaging models that are multimodal, multi-task, explainable and that generalize across hospitals.
Read more → -

Sketch-based Understanding
Sketching is one of the most natural ways people communicate. We build encoders that understand sketches and connect them with photos: for retrieval, localization, segmentation and generation.
Read more → -

Efficient One-Shot Detection
Detecting objects from classes the model has never seen, given a single example — even when that example is a sketch.
Read more → -

Generative Models
Conditional generative models — from GANs to diffusion — that turn a sketch into a realistic image, or a low-dose CT into a diagnostic-quality one.
Read more → -

Video Analysis
Models that describe video in the wild — news, sports, streaming — and that let you search video catalogues with a sketch.
Read more →