Video Quality Assessment for Resolution Cross-Over in Live Sports

2025-04-01

AI Summary

In adaptive bitrate (ABR) live streaming, inaccurate prediction of resolution cross-over points—where perceived quality reverses between two resolutions—hampers optimal bitrate adaptation. This paper identifies a systematic bias in the widely adopted Absolute Category Rating (ACR) subjective evaluation paradigm when modeling such cross-overs. Method: We propose a more robust Pairwise Comparison (PC) subjective assessment paradigm and empirically demonstrate, for the first time, its superior accuracy in cross-over localization. Building on PC, we design Resolution Cross-over Quality Loss (RCQL), a novel quality metric specifically tailored to resolution switching. We further introduce LSCO, the first large-scale subjective dataset focused on live sports streaming. Results: RCQL significantly improves cross-over prediction accuracy; on the LSCO benchmark, state-of-the-art VQMs (e.g., VMAF) incur up to 37% cross-over error, whereas the PC+RCQL framework reduces localization error by 52%.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Evaluation and Analysis
• Computer Vision: Video Understanding & Activity Analysis

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success
• Security and Privacy: Large-scale security measurements

Abstract

In adaptive bitrate streaming, resolution cross-over refers to the point on the convex hull where the encoding resolution should switch to achieve better quality. Accurate cross-over prediction is crucial for streaming providers to optimize resolution at given bandwidths. Most existing works rely on objective Video Quality Metrics (VQM), particularly VMAF, to determine the resolution cross-over. However, these metrics have limitations in accurately predicting resolution cross-overs. Furthermore, widely used VQMs are often trained on subjective datasets collected using the Absolute Category Rating (ACR) methodologies, which we demonstrate introduces significant uncertainty and errors in resolution cross-over predictions. To address these problems, we first investigate different subjective methodologies and demonstrate that Pairwise Comparison (PC) achieves better cross-over accuracy than ACR. We then propose a novel metric, Resolution Cross-over Quality Loss (RCQL), to measure the quality loss caused by resolution cross-over errors. Furthermore, we collected a new subjective dataset (LSCO) focusing on live streaming scenarios and evaluated widely used VQMs, by benchmarking their resolution cross-over accuracy.

Problem

• Improving resolution cross-over prediction in adaptive streaming
• Addressing limitations of current Video Quality Metrics (VQM)
• Evaluating subjective methodologies for accurate quality assessment

Innovation

• Pairwise Comparison improves cross-over accuracy
• Proposed RCQL metric measures quality loss
• New LSCO dataset evaluates VQMs accuracy