DINOv2 Resolution-Prediction Network Overview

Paper context

Paper title: Seeing enough: non-reference perceptual resolution selection for power-efficient client-side rendering Abstract: Many client-side applications, especially games, render video at high resolution and frame rate on power-constrained devices, even when users perceive little or no benefit from all those extra pixels. Existing perceptual video quality metrics can indicate when a lower resolution is "good enough", but they are full-reference and computationally expensive, making them impractical for real-world applications and deployment on-device. In this work, we leverage the spatio-temporal limits of the human visual system and propose a non-reference method that predicts, from the rendered video alone, the lowest resolution that remains perceptually indistinguishable from the best available option, enabling power-efficient client-side rendering. Our approach is codec-agnostic and requires only minimal modifications to existing infrastructure. The network is trained on a large dataset of rendered content labeled with a full-reference perceptual video quality metric. The prediction significantly enhances perceptual quality while substantially reducing computational costs, suggesting a practical path toward perception-guided, power-efficient client-side rendering. Passages referencing this figure: nostic and requires only minimal modifications to existing infrastructure. The network is trained on a large dataset of rendered content labeled with a full-reference perceptual video quality metric. The prediction significantly enhances perceptual quality while substantially reducing computational costs, suggesting a practical path toward perception-guided, power-efficient client-side rendering. Figure 1 : Client devices render interactive content at a high frame rate (120 Hz). For each 250 ms clip, we extract motion vectors and a short rendered image sequence; our resolution predictor encodes the relevant spatial and temporal structures that give rise to visible distortions. These motion and appearance features are fused by a non-reference resolution predictor, which estimates the next p