Multimodal learning · Computer vision · Medical Imaging

Connecting modalities to build more reliable medical imaging AI.

I study multimodal learning for medical imaging AI: how models can combine different imaging modalities and complementary sources of information to form reliable representations and predictions. My broader work spans computer vision, crowd counting, vision–language learning, and human-centred perception. Here, I explain each project from its question and intuition to its evidence and limitations.

Explore my research Read my posts

“Do not block the way of inquiry.”
— Charles S. Peirce, The First Rule of Logic ↗

01 / Selected research

Ideas, evidence, and limits.

3 selected works; 6 publications in the complete archive.

View all publications

31 Jul 2025Preprint

ZIP: Scalable Crowd Counting via Zero-Inflated Poisson Modeling

When more than 95% of local crowd-map blocks are empty, is squared-error regression still the right statistical model?

Central idea

ZIP predicts two quantities per block: a structural-zero probability for background, body parts, and other non-head-centre regions, and a Poisson rate for candidate head-centre regions. Integer count bins stabilize rate prediction, while the joint likelihood learns the distinction between structural zeros and sampling zeros directly from point annotations.

Task · Crowd countingMethod · Probabilistic modelsGoal · Scalability

Full explanation Paper Code Demo

14 Mar 2024ICME 2025

CLIP-EBC: CLIP Can Count Accurately through Enhanced Blockwise Classification

CLIP was trained to match images and words, not to regress continuous values. Can counting be reformulated so that its recognition knowledge remains useful?

Task · Crowd countingMethod · Vision–languageGoal · Accuracy

Read explanation

13 Apr 2023CVPRW 2023

Robust Multiview Multimodal Driver Monitoring System Using Masked Multi-Head Self-Attention

How should four heterogeneous video sources be fused, and can the same model remain useful when one or more sources fail?

Task · Human behaviourMethod · Multimodal fusionGoal · Robustness

Read explanation

02 / My posts

Notes that make research legible.

The first full post is now available in matching English and Chinese editions.

Rereading the Transformer Upgrade Path · 20 min read

Rereading the Transformer Upgrade Path (1): Tracing Sinusoidal Position Encoding to Its Source

View all posts

03 / About

“False facts are highly injurious to the progress of science, for they often endure long.”

— Charles Darwin, The Descent of Man ↗

Research is more than a list of results. I use this site to connect the problems I choose, the models I build, the evidence I trust, and the limits I learn from. My current focus is multimodal learning for medical imaging AI, building on earlier work in crowd counting, vision–language models, multimodal driver monitoring, and human-centred vision.

Full profile English CV Email me