New AI system adapts frame rate and resolution for smarter game streaming
Neural net predicts optimal settings to cut bandwidth costs while boosting perceived quality.
Deep Dive
Yaru Liu, Joseph G. March, and Rafal K. Mantiuk propose a system that uses a lightweight neural network to adaptively adjust frame rate and resolution for streaming rendered content. It exploits spatio‑temporal limits of human vision, is codec‑agnostic, and requires minimal modifications to existing rendering infrastructure. The model is trained on a large dataset labeled with a perceptual video quality metric. The result: better perceived quality under bandwidth constraints.
Key Points
- Lightweight neural network predicts optimal frame rate + resolution per frame based on bandwidth, content, and motion velocity.
- Codec‑agnostic design requires only minimal modifications to existing rendering infrastructure.
- Trained on a large dataset of rendered content labeled with a perceptual video quality metric.
Why It Matters
Enables higher perceived quality for cloud gaming and VR on constrained networks without expensive infrastructure overhauls.