Coarse-to-Real framework generates realistic urban crowd videos from 3D simulations

Researchers have developed a new generative rendering framework called C2R (Coarse-to-Real) designed to synthesize realistic urban crowd videos from simplified 3D simulations. This system leverages coarse 3D models for scene layout and motion control, while a neural renderer generates realistic appearance and dynamics guided by text prompts. To address the lack of paired training data, C2R employs a two-stage domain-hedging strategy, first learning from extensive real footage and then fine-tuning with limited synthetic data to ensure controllability and temporal consistency. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a novel approach for generating controllable, realistic videos from simplified 3D inputs, potentially impacting game development and virtual scene creation.

RANK_REASON This is a research paper detailing a new generative rendering framework.

Read on arXiv cs.CV →

• Coarse-to-Real
• C2R
• arXiv
• Hugging Face

• paper
• other