CAST: An Intricate-Scene Aware Adaptive Bitrate Approach for Video Streaming via Parallel Training

Weihe Li, Jiawei Huang, Yu Liang, Jingling Liu, Wenlu Zhang, Wenjun Lyu, Jianxin Wang

Abstract: Adaptive Bitrate (ABR) algorithms have become increasingly important for delivering high-quality video content over fluctuating networks. Considering the complexity of video scenes, video chunks can be separated into two categories: those with intricate scenes and those with simple scenes. In practice, improving the quality of intricate chunks can lead to more significant improvements in Quality of Experience (QoE) than improving simple chunks. However, current schemes either assign equal priority to all chunks or optimize using a fixed linear-based reward function, making them inadequate for meeting real-world requirements. To tackle these limitations, this paper introduces a novel ABR approach that explicitly considers bitrate adaptation as the primary objective. The proposed approach, CAST (Complex-scene Aware bitrate algorithm via Self-play reinforcemenT learning), leverages the power of parallel computing with multiple agents to train a neural network, aiming to achieve superior video playback quality for intricate scenes while minimizing frequent freezing events. The extensive trace-driven evaluation and subjective test results demonstrate that CAST outperforms existing off-the-shelf schemes.