Reproduce “JEPA Guided Diffusion: Predictive Vision-Language Conditioning for Generative Traffic Forecasting” 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
Trinh Tra Giang Nguyen, Thanh Nguyen Vo, Nguyen Hoai Thuong Bui, Ha Duc Bui
JEPA Guided Diffusion forecasts traffic by combining V-JEPA 2.1 features from observed video with behavioral captions, then predicting conditioning tokens for a separate, frozen Cosmos-Predict2.5 generator. In a controlled no-postprocessing comparison, it scores 73.5120 versus 72.0975 for a VLM-guided baseline. It ranks third on AI City Challenge 2026 Track 5, not a safety test.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.