CTC: Content-Aware Tailoring of Adaptive Video Streaming using Multi-Head Critic Network

In this paper, we aim to enhance video streaming quality by taking into account a simple observation: users tend to focus on specific areas within a video. For instance, low-quality or stall events during scoring moments in sports videos can lead to user frustration. However, most existing video streaming solutions treat all scenes equally. In our work, we introduce CTC, an ABR algorithm that adjusts its policy based on scenes. To achieve this, we first model dynamic QoE based on scenes and then use reinforcement learning to adapt the policy in real-time. As a result, CTC significantly improves QoE by adjusting its policy according to content compared to existing work.