Quality-Aware Dynamic Resolution Adaptation Framework for Adaptive Video Streaming

TL;DR

QADRA presents a latency-aware, quality-driven framework for adaptive video streaming that predicts $XPSNR$ for VVC-coded representations from spatiotemporal features and encoding settings using $XGBoost$, then optimizes per-segment encoding resolution and $QP$ under encoding/decoding time constraints. It introduces a convex-hull online resolution selection, a two-model $QP$ predictor with linear bitrate mapping, and a JND-based representation elimination to prune the bitrate ladder, all implemented in an open-source Python framework. Empirical results show that QADRA can achieve higher RD performance (higher $XPSNR$ at a given bitrate) than standard ladders while exposing the trade-off between quality and encoding/decoding time, illustrating practical gains for energy-efficient, low-latency streaming. The approach enables scalable, content-adaptive bitrate ladders that adapt to device and network diversity, with potential extensions to VR/AR and distributed encoding workflows.

Abstract

Traditional per-title encoding schemes aim to optimize encoding resolutions to deliver the highest perceptual quality for each representation. XPSNR is observed to correlate better with the subjective quality of VVC-coded bitstreams. Towards this realization, we predict the average XPSNR of VVC-coded bitstreams using spatiotemporal complexity features of the video and the target encoding configuration using an XGBoost-based model. Based on the predicted XPSNR scores, we introduce a Quality-A ware Dynamic Resolution Adaptation (QADRA) framework for adaptive video streaming applications, where we determine the convex-hull online. Furthermore, keeping the encoding and decoding times within an acceptable threshold is mandatory for smooth and energy-efficient streaming. Hence, QADRA determines the encoding resolution and quantization parameter (QP) for each target bitrate by maximizing XPSNR while constraining the maximum encoding and/ or decoding time below a threshold. QADRA implements a JND-based representation elimination algorithm to remove perceptually redundant representations from the bitrate ladder. QADRA is an open-source Python-based framework published under the GNU GPLv3 license. Github: https://github.com/PhoenixVideo/QADRA Online documentation: https://phoenixvideo.github.io/QADRA/

Figures (4)

• Figure 1: Rate-distortion (RD) and rate-encoding time curves of representative sequences (segments) of Inter-4K dataset stergiou_adapool_2023 encoded at 360p, 720p, 1080p and 2160p resolutions using VVenC encoder wieckowski_vvenc_2021 at faster preset. Here, XPSNR helmrich_xpsnr_2020 is used as the quality metric.
• Figure 2: Encoding using QADRA framework envisioned in this paper.
• Figure 3: Selected encoding resolutions of representative video sequences (segments) using QADRA ($r_{\text{max}}=2160$).
• Figure 4: RD curves, and encoding times of representative video sequences (segments) using QADRA ($r_{\text{max}}=2160$).