Reproduce “EyeWorld: A Generative World Model of Ocular State and Dynamics” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ziyu Gao, Xinyuan Wu, Xiaolan Chen, Zhuoran Liu, Ruoyu Chen, Bowen Liu, Bingjie Yan, Zhenhan Wang and 5 more
Researchers at The Hong Kong Polytechnic University developed EyeWorld, a generative world model that learns an observation-stable latent ocular state across multiple imaging modalities and models time-conditioned state transitions for ophthalmic disease. The model outperforms existing baselines across tasks like anatomical parsing, cross-modality translation, image quality enhancement, and disease progression prediction.
Key insights
- A unified, observation-stable latent ocular state can robustly preserve lesion-level semantics and boundary-sensitive structures across diverse imaging modalities, devices, and quality variations.
- Explicitly modeling time-conditioned state transitions enables prognosis-oriented simulation of disease progression, forecasting morphological evolution while maintaining global retinal geometry.
- Exemplar-based conditioning offers a flexible interface to infer and apply transformation rules from input-output examples across various tasks and modalities, reducing reliance on precise prompt phrasing.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.