New framework CodecArena assesses video codec quality using visual reinforcement learning

Researchers have introduced CodecArena, a novel vision-language framework designed to assess video codec quality, particularly in low and ultra-low bitrate scenarios. Unlike existing metrics that focus on feature similarity, CodecArena evaluates content fidelity by comparing reference videos with their reconstructions, considering aspects like identity, objects, text, texture, and temporal consistency. The framework is optimized using Facet-GRPO, a visual reinforcement learning method, and is supported by two new resources: CodecArena-1K for automated preference dataset generation and CodecArena-Bench for human-ranked, out-of-domain evaluation. AI

IMPACT This new framework could lead to more accurate and interpretable evaluations of video compression techniques, potentially improving the efficiency and quality of video streaming and generation.

RANK_REASON The item is an academic paper detailing a new framework and dataset for video codec quality assessment. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• CodecArena
• CodecArena-1K
• CodecArena-Bench
• Facet-GRPO
• LPIPS

• paper
• other

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