Streaming systems that employ video super-resolution (SR) often rely on a single, generic neural network model for all content types, resulting in suboptimal visual quality across diverse scenes. To address this limitation, we propose a metadata-guided hot swapping mechanism that enables the dynamic selection of specialized, fine-tuned SR models during streaming. The system uses content type change signals transmitted via an auxiliary metadata track, which is prioritized over media tracks to ensure early arrival. This allows the client to preload the appropriate neural network before the content type changes, minimizing startup delay and improving responsiveness. The end-to-end workflow includes content detection through source mapping, high-priority metadata transmission, message extraction, neural network preloading and SR application.

The increasing demand for real-time communication (RTC) applications necessitates robust and reliable systems. Seamless media delivery depends on an accurate assessment of the network conditions, with bandwidth estimation (BWE) being crucial for maintaining system reliability and achieving good quality of experience (QoE) for the users. BWE poses a significant challenge due to dynamic network conditions, limited information availability and computational complexity. The Second Bandwidth Estimation Challenge, organized within ACM MMSys 2024, aims to enhance RTC user QoE by developing a deep learning-based bandwidth estimator using offline reinforcement learning. This paper presents our solution, ranked second in the grand challenge. This solution employs an actor-critic approach to achieve accurate real-time BWE by relying solely on observed network statistics. Due to the offline setting of the challenge, the critic network is trained separately from the actor network to estimate the action quality without interacting with the real environment. Furthermore, the quality prediction by the critic is adjusted by a predefined conservation factor to address overshooting the bandwidth values. The solution's source code is publicly available at https://github.com/streaming-university/FARC.

Recent years have witnessed video streaming demands evolve into one of the most popular Internet applications. With the ever-increasing personalized demands for highdefinition and low-latency video streaming services, networkassisted video streaming schemes employing modern networking paradigms have become a promising complementary solution in the HTTP Adaptive Streaming (HAS) context. The emergence of such techniques addresses long-standing challenges of enhancing users’ Quality of Experience (QoE), end-to-end (E2E) latency, as well as network utilization. However, designing a cost-effective, scalable, and flexible network-assisted video streaming architecture that supports the aforementioned requirements for live streaming services is still an open challenge. This article leverages novel networking paradigms, i.e., edge computing and Network Function Virtualization (NFV), and promising video solutions, i.e., HAS, Video Super-Resolution (SR), and Distributed Video Transcoding (TR), to introduce A Latency-and cost-aware hybrId P2P-CDN framework for liVe video strEaming (ALIVE). We first introduce the ALIVE multi-layer architecture and design an action tree that considers all feasible resources (i.e., storage, computation, and bandwidth) provided by peers, edge, and CDN servers for serving peer requests with acceptable latency and quality. We then formulate the problem as a Mixed Integer Linear Programming (MILP) optimization model executed at the edge of the network. To alleviate the optimization model’s high time complexity, we propose a lightweight heuristic, namely, Greedy-Based Algorithm (GBA). Finally, we (i) design and instantiate a large-scale cloud-based testbed including 350 HAS players, (ii) deploy ALIVE on it, and (iii) conduct a series of experiments to evaluate the performance of ALIVE in various scenarios. Experimental results indicate that ALIVE (i) improves the users’ QoE by at least 22%, (ii) decreases incurred cost of the streaming service provider by at least 34%, (iii) shortens clients’ serving latency by at least 40%, (iv) enhances edge server energy consumption by at least 31%, and (v) reduces backhaul bandwidth usage by at least 24% compared to baseline approaches.

Designing a cost-effective, scalable, and flexible architecture that supports low latency and high quality live video streaming is still a challenge for Over-The-Top (OTT) service providers. To cope with this issue, this paper leverages Peer-to-Peer (P2P), Content Delivery Network (CDN), edge computing, Network Function Virtualization (NFV), and distributed video transcoding paradigms to introduce a hybRId P2P-DN arcHfiTecture for livE video stReaming (RICHTER). We first introduce RICHTER's multi-layer architecture and design an action tree that considers all feasible resources provided by peers, edge, and CDN servers for serving peer requests with minimum latency and maximum quality. We then formulate the problem as an optimization model executed at the edge of the network. We present an Online Learning (OL) approach that leverages an unsupervised Self Organizing Map (SOM) to (i) alleviate the time complexity issue of the optimization model and (ii) make it a suitable solution for large-scale scenarios, by enabling decisions for groups of requests instead of for single requests. Finally, we implement the RICHTER framework, conduct our experiments on a large-scale cloud-based testbed including 350 HAS players, and compare its effectiveness with baseline systems. The experimental results illustrate that RICHTER outperforms baseline schemes in terms of users' Quality of Experience (QoE), latency, and network utilization, by at least 59%, 39%, and 70% respectively.

Recently, mobile devices have become paramount in online video streaming. Adaptive bitrate (ABR) algorithms of players responsible for selecting the quality of the videos face critical challenges in providing a high Quality of Experience (QoE) for end users. One open issue is how to ensure the optimal experience for heterogeneous devices in the context of extreme variation of mobile broadband networks. Additionally, end users may have different priorities on video quality and data usage (i.e., the amount of data downloaded to the devices through the mobile networks). A generic mechanism for players that enables specification of various policies to meet end users’ needs is still missing. In this paper, we propose a weighted sum model, namely WISH, that yields high QoE of the video and allows end users to express their preferences among different parameters (i.e., data usage, stall events, and video quality) of video streaming. WISH has been implemented into ExoPlayer, a popular player used in many mobile applications. The experimental results show that WISH improves the QoE by up to 17.6% while saving 36.4% of data usage compared to state-of-the-art ABR algorithms and provides dynamic adaptation to end users’ requirements.

Video traffic comprises the majority of today's Internet traffic, and HTTP Adaptive Streaming (HAS) is the preferred method to deliver video content over the Internet. Increasing demand for video and the improvements in the video display conditions over the years caused an increase in the video coding complexity. This increased complexity brought the need for more efficient video streaming and coding solutions. The latest standard video codecs can reduce the size of the videos by using more efficient tools with higher time-complexities. The plans for integrating machine learning into upcoming video codecs raised the interest in applied machine learning for video coding. In this doctoral study, we aim to propose applied machine learning methods to video coding, focusing on HTTP adaptive streaming. We present four primary research questions to target different challenges in video coding for HTTP adaptive streaming.

Adaptive HTTP streaming provides multiple representations of the same content at different bit-rates and resolutions and allows the client to request segments from the available representations in a dynamic, adaptive way depending on its context. The growing number of representations in adaptive HTTP streaming makes encoding of one video segment at different representations a challenging task in terms of encoding time-complexity. In this paper, information of both highest and lowest quality representations are used to limit Rate Distortion Optimization (RDO) process for each Coding Unit Tree (CTU) in High Efficiency Video Coding. Our proposed method first encodes the highest quality representation and consequently uses its information to encode the lowest quality representation. Thereafter, information from both the highest and the lowest quality representations are used to predict features of intermediate quality representations. In particular, the block structure and the selected reference frame of both highest and lowest quality representations are used to predict and shorten the RDO process of each CTU for intermediate quality representations. Our proposed method introduces a delay of two CTUs if parallel encoding is used. Experimental results show significant reduction in time-complexity over the reference software (38%) and the state-of-the-art (10%) while quality degradation is negligible.

HTTP Adaptive Streaming (HAS) is the most common approach for delivering video content over the Internet. The requirement to encode the same content at different quality levels (i.e., representations) in HAS is a challenging problem for content providers. Fast multirate encoding approaches try to accelerate this process by reusing information from previously encoded representations. In this paper, we propose to use convolutional neural networks (CNNs) to speed up the encoding of multiple representations with a specific focus on parallel encoding. In parallel encoding, the overall time-complexity is limited to the maximum time-complexity of one of the representations that are encoded in parallel. Therefore, instead of reducing the time-complexity for all representations, the highest time-complexities are reduced. Experimental results show that FaME-ML achieves significant time-complexity savings in parallel encoding scenarios (41% in average) with a slight increase in bitrate and quality degradation compared to the HEVC reference software.