QARC MEETING SIMULATOR
So, our simulator should simulate the process of the packets coming and leaving in different network conditions, and keep track of the timestamps, by which we can get the corresponding queuing delay. To train our model, our training data should consist of queuing delay rather than one-way delay.
The inputs of the VQRL are past time network status observed and future video quality predicted by VQPN, and the output is the bitrate for the next video with high video quality and low latency. VQRL uses A3C (Mnih et al., 2016), a DRL method, to train the neural network. One is Video Quality Prediction Network (VQPN), which can predict future video quality via previous video frames the other is Video Quality Reinforcement Learning (VQRL). To overcome this, we meticulously divide this complexed RL model into two feasible and useful models: However, if we directly import raw pictures as the inputs of state, the state space will cause “state explosion” (Clarke et al., 2012). In detail, QARC uses DRL method to train a neural network to select the bitrate for future video frames based on past time network status observed and historical video frames. In this paper, we propose QARC(video Quality Awareness Rate Control), a novel deep-learning based rate control algorithm aiming to obtain high video quality and low latency.ĭue to that fixed rules fail to effectively handle the complicated scenarios caused by perplexing network conditions and various video content, we leverage DRL-based method to select the future video bitrate, which can adjust itself automatically to the variety of its inputs. For example, if a video footage consists of darkness and few objects, a low bitrate may also provide a barely satisfactory perceptual video quality but can save large bandwidth resources, and the example is shown in Figure 1(a). However, due to the inequality between high video quality and high bitrate, this strategy may cause a large waste of bandwidth resources. The same strategy of them is to select bitrate as high as possible with the permission of network condition. Peterson, 1995), LEDBAT (Over UDP) (RossiĮt al., 2010)), and model-based approach (Google Congestion Control(GCC) (Carlucci et al., 2016), Rebera (Kurdoglu et al., 2016)).
Many rate control approaches have been proposed to tackle the problem, such as loss-based approach (TFRC (HandleyĮt al., 1999)), delay-based approach (Vegas (Brakmo and Due to the complicated environment and stochastic property in various network conditions, transmitting video stream with high video bitrate and low latency has become the fundamental challenge in real-time video streaming scenario. Live video streams are being published and watched by different applications(e.g., Twitch, Kwai, Douyu) at any time, from anywhere, and under any network environments. Recent years have witnessed a rapid increase in the requirements of real-time video streaming (Cisco, 2017). High bitrate method on various network conditions also yields a solid result. With improvements in average video quality of 18% - 25% and decreases in WeĮvaluate QARC over a trace-driven emulation, outperforming existing approach Problem", we design a neural network to predict future perceptual video qualityĪs a vector for taking the place of the raw picture in the DRL's inputs. Train a neural network to select future bitrates based on previously observed
Starting from scratch, QARC uses deep reinforcement learning(DRL) algorithm to Video quality with possibly lower sending rate and transmission latency. Rate Control), a rate control algorithm that aims to have a higher perceptual In this paper, we propose QARC (video Quality Awareness
QARC MEETING HOW TO
Sending bitrate and video quality, which motivates us to focus on how to get aīalance between them. Nevertheless, we notice that there exists a trade-off between Methods have been proposed to provide high video bitrates instead of video To tackle this problem, most adaptive bitrate control Due to the fluctuation of throughput under various networkĬonditions, how to choose a proper bitrate adaptively has become an upcomingĪnd interestingly issue. Real-time video streaming is now one of the main applications in all networkĮnvironments.
0 kommentar(er)
