RealWeather model translates weather in driving videos with scene fidelity

Researchers have developed RealWeather, a novel driving world model designed for realistic and scene-faithful weather translation. This model learns weather dynamics directly from real-world videos, employing an iterative data-refinement strategy called Progressive Realism Bootstrapping. To ensure structural integrity and prevent hallucinations, RealWeather incorporates Scene-Fidelity RL Optimization, a reward-driven policy that penalizes alterations to critical driving elements. Experiments show RealWeather surpasses existing methods in visual realism and structural preservation, enabling the generation of long-tail weather scenarios and robust out-of-distribution generalization. AI

IMPACT Enhances realism and scene preservation in synthetic driving data, potentially improving autonomous vehicle training and testing.

RANK_REASON This is a research paper detailing a new method for weather translation in driving videos. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Progressive Realism Bootstrapping
• Pseudo-Clear Generation
• RealWeather
• Scene-Fidelity RL Optimization

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →