New PW-VQM metric beats SSIM and VMAF on sports video quality
0.9511 Spearman correlation outperforms VMAF, SSIM, LPIPS, and FUNQUE on sports content
Deep Dive
A new full-reference video quality metric called PW-VQM accounts for unequal spatial attention in sports videos. By weighting foreground regions identified via object detection and optical flow, PW-VQM achieves a Spearman correlation of 0.9511, surpassing SSIM, VMAF, FUNQUE, and LPIPS. The metric is designed for asymmetric encoded content, enabling bitrate savings by allocating more bits to salient regions.
Key Points
- PW-VQM uses object detection and optical flow to identify foreground regions for higher perceptual weighting
- Achieves 0.9511 Spearman correlation, beating VMAF (0.925) and FUNQUE (0.938) on sports video
- Enables bitrate savings by allocating more bits to salient areas in asymmetric encoding
Why It Matters
Streaming services can cut bandwidth on sports by focusing bits where viewers actually look.