Dynamic point clouds enable immersive six-degrees-of-freedom experiences but require complex content preparation pipelines to support efficient adaptive delivery at scale. In practice, workflows for dynamic point cloud streaming remain fragmented: spatial tiling, encoding, segmentation, and adaptive streaming signaling are often handled by disconnected tools, ad hoc scripts, or platform-dependent binaries that are difficult to obtain, compile, and reproduce, particularly in Linux-based research environments. This paper presents V3CTK, an end-to-end content preparation toolkit for tiled dynamic point cloud streaming that automates these steps with minimal manual intervention. V3CTK unifies segment-adaptive spatial tiling, per-tile sequence-level V-PCC encoding, GoF-aligned V3C segmentation, and MPEG-DASH MPD generation within a single modular pipeline, while still allowing individual stages to be executed independently when needed. The generated bitstream segments and manifests follow MPEG-I V3C/V-PCC (ISO/IEC 23090-5) and MPEG-DASH (ISO/IEC 23009-1) signaling conventions, while incorporating explicit extensions to support tiled and component-level streaming. In addition to unified V3C bitstreams, the pipeline supports component-wise bitstream generation with multiple representations per component, and provides a lightweight multiplexing step to reconstruct a fully decodable unified V3C bitstream from per-component streams prior to playback. V3CTK exposes finegrained parameter control through command-line and web-based interfaces, offering a unified workflow for the research community and simplifying setup through Dockerized encoding.

Volumetric media streaming is a key enabler of immersive 6DoF applications such as virtual reality and interactive telepresence. Dynamic point clouds, standardized by MPEG through Video-based Point Cloud Compression (V-PCC), offer an efficient representation for such content. However, existing datasets and quality evaluation studies are largely human-centric and fail to capture the heterogeneity of real-world volumetric scenes in terms of object category, motion dynamics, and geometric complexity, limiting content-aware systems analysis. In this paper, we introduce QUEST-PCC1, an open-source reference dataset for objective quality and rate-distortion analysis of heterogeneous dynamic point clouds encoded using the MPEG V-PCC standard. QUEST-PCC spans six object categories, including articulated, rigid, and non-human objects, and covers a wide range of motion and structural complexity. The dataset provides reconstructed point clouds, compressed V-PCC bitstreams, and standardized per-frame MPEG PCC quality metrics across multiple rate levels, yielding content-dependent bitrate ladders that span from sub-1 Mbps to approximately 80 Mbps. Through systematic sequence-level and category-level analysis, we show that encoding bitrate and reconstruction quality are governed primarily by dynamic surface topology and structural complexity rather than point count alone, revealing content-dependent rate-distortion behavior that is not captured by existing human-centric datasets and is critical for content-aware compression and adaptive volumetric streaming.

Multi-CDN strategies are crucial for ensuring high Quality of Experience (QoE) in adaptive video streaming. Yet existing approaches-whether heuristic or learning-based-largely depend on aggregated client metrics, leading to coarse-grained CDN selection that constrains QoE and neglects operational costs. To address these shortcomings, we propose CADENCE, a multi-agent reinforcement learning framework built on Centralized Training with Decentralized Execution (CTDE). CADENCE deploys per-client agents to perform fine-grained, real-time CDN selection tailored to each streaming session. Unlike prior work, CADENCE jointly optimizes client-side QoE and content provider-side costs while proactively mitigating CDN overload. Through extensive high-idelity trace-driven emulation, we demonstrate that CADENCE signiicantly outperforms state-of-the-art baselines, improving average VMAF by up to 21% with minimal rebuffering and reducing operational costs by 35%.

Volumetric media in the form of dynamic point clouds is increasingly adopted for immersive applications such as virtual reality, telepresence, digital twins, and interactive 3D experiences, yet practical support for browser-native visualization remains limited. Existing solutions often rely on heavyweight native players, offline preprocessing, or specialized plugins, creating a significant barrier to accessibility and deployment at scale. In this demo, we present NAVIS, a fully browser-based volumetric media player that enables interactive playback and exploration of dynamic PLY point-cloud sequences using standard web technologies. NAVIS combines WebGL-based rendering with a parallel web-worker parsing pipeline, buffered playback with adaptive concurrency control, and interactive navigation modes, allowing large volumetric datasets to be visualized smoothly directly within the browser. We instrument NAVIS to expose key performance metrics, including startup delay, parsing latency, and stall duration, and evaluate it across multiple volumetric datasets at 30 fps and 60 fps. Our results show that increasing worker parallelism substantially reduces startup delay and mitigates playback stalls, demonstrating that high-performance volumetric media visualization is feasible on the web without native dependencies. NAVIS provides a practical and extensible foundation for future browser-native immersive media systems.

Forward Error Correction (FEC) remains essential for protecting video streaming against packet loss, yet most real deployments still rely on static, coarse-grained configurations that cannot react to rapid shifts in loss rate, goodput, or client buffer levels. These rigid settings often create inefficiencies: unnecessary redundancy that suppresses throughput during stable periods, and insufficient protection during bursty losses, especially when shallow buffers and oversized blocks increase stall risk. To address these challenges, we present TAROT, a cross-layer, optimization-driven FEC controller that selects redundancy, block size, and symbolization on a per-segment basis. TAROT is codec-agnostic--supporting Reed-Solomon, RaptorQ, and XOR-based codes--and evaluates a pre-computed candidate set using a fine-grained scoring model. The scoring function jointly incorporates transport-layer loss and goodput, application layer buffer dynamics, and block-level timing constraints to penalize insufficient coverage, excessive overhead, and slow block completion. To enable realistic testing, we extend the SABRE simulator 1 with two new modules: a high-fidelity packet-loss generator that replays diverse multi-trace loss patterns, and a modular FEC benchmarking layer supporting arbitrary code/parameter combinations. Across Low-Latency Live (LLL) and Video-on-Demand (VoD) streaming modes, diverse network traces, and multiple ABR algorithms, TAROT reduces FEC overhead by up to 43% while improving perceptual quality by 10 VMAF units with minimal rebuffering, achieving a stronger overhead-quality balance than static FECs.

The ETSI TS 103 998 content steering standard enables runtime switching across multiple Content Delivery Networks (CDNs), but current deployments still rely on static heuristics that struggle under dynamic and low-latency network conditions. In this demo, we present the integration of CADENCE, a fine-grained multi-agent multi-CDN controller, directly into the Dash.js reference player, transforming it into a practical experimentation platform for CDN steering. The enhanced player exposes real-time CDN telemetry, per-CDN QoE scoring, and runtime strategy selection, enabling transparent and high-fidelity comparison of steering approaches under identical playback conditions. We additionally include an optional short-horizon forecasting module as a proof-of-concept extension and use controlled ablation experiments to show how forecasted telemetry can mitigate the impact of delayed feedback with negligible overhead. Overall, the demo illustrates how CADENCE —and more broadly, learning-based or multi-agent steering controllers—can be seamlessly embedded into Dash.js to study robust multi-CDN behavior in realistic deployments.

The QUIC transport protocol represents a significant evolution in web transport technologies, offering improved performance and reduced latency compared to traditional protocols like TCP. Given the growing number of QUIC implementations, understanding their performance, particularly in video streaming contexts, is essential. This paper presents a comprehensive analysis of various QUIC implementations, focusing on their transport-layer congestion control (CC) performance and its impact on HTTP Adaptive Streaming (HAS) in single-server, multi-client environments. Through extensive trace-driven experiments, we explore how different QUIC CCs impact adaptive bitrate (ABR) algorithms in two video streaming scenarios: video-on-demand (VoD) and low-latency live streaming (LLL). Our study aims to shed light on the impact of QUIC CC implementations, queuing strategies, and cooperative versus competitive dynamics of QUIC streams on user QoE under diverse network conditions. Our results demonstrate that identical CC algorithms across different QUIC implementations can lead to significant performance variations, directly impacting the QoE of video streaming sessions. These findings offer valuable insights into the effectiveness of various QUIC implementations and their implications for optimizing QoE, underscoring the need for intelligent cross-layer designs that integrate QUIC CC and ABR schemes to enhance overall streaming performance.

The rapid adoption of QUIC as a transport protocol has transformed content delivery by reducing latency, enhancing congestion control (CC), and enabling more efficient multiplexing. With the advent of 5G networks, which support ultra-low latency and high bandwidth, streaming high-resolution video at 4K and beyond has become increasingly viable. However, optimizing Quality of Experience (QoE) in mobile networks remains challenging due to the complex interactions among Adaptive Bit Rate (ABR) schemes at the application layer, CC algorithms at the transport layer, and Radio Link Control (RLC) queuing at the link layer in the 5G network. While prior studies have largely examined these components in isolation, this work presents a comprehensive analysis of the impact of modern active queue management (AQM) strategies, such as RED and L4S, on video streaming over diverse QUIC implementations—focusing particularly on their interaction with the RLC buffer in 5G environments and the interplay between CC algorithms and ABR schemes. Our findings demonstrate that the effectiveness of AQM strategies in improving video streaming QoE is intrinsically linked to their dynamic interaction with QUIC implementations, CC algorithms and ABR schemes—highlighting that isolated optimizations are insufficient. This intricate interdependence necessitates holistic, cross-layer adaptive mechanisms capable of real-time coordination between network, transport and application layers, which are crucial for leveraging the capabilities of 5G networks to deliver robust, adaptive, and high-quality video.

The increasing demand for high-quality real-time 360° video streams from mobile platforms, such as 5G-connected Unmanned Aerial Vehicles (UAVs), is challenging modern B5G networks. Vehicular mobility and fluctuating conditions in high-altitude, high-speed scenarios, known as high volatility, complicate maintaining an effective Quality of Experience (QoE) for cellular networks. This work introduces FlyBit, a Deep Reinforcement Learning (DRL)-based bitrate selection framework for live 360° video streaming in 5G-connected UAV applications, designed to enhance video quality, reduce packet loss, and minimize End-to-End (E2E) latency. We developed and deployed a real-world testbed to evaluate the impact of dynamic network conditions, UAV mobility, and trajectory on streaming performance, analyzing FlyBit with real-world data. Experimental results show that FlyBit improves Video Multimethod Assessment Fusion (VMAF) by ~29% and average bitrate by ~50%, while maintaining low latency and packet loss compared to baseline approaches, demonstrating its ability to adjust bitrate in real-time and significantly improve QoE for ultra-low-latency video streaming.

Multi-CDN strategies have become increasingly important in enhancing Quality of Experience (QoE) for adaptive video streaming. The recent development of the content steering standard (ETSI TS 103 998) aims to facilitate real-time decision-making about the best-performing CDN by gathering statistics from both players and CDNs. However, this task presents significant challenges. Existing solutions for CDN selection rely on heuristics, which often fail to adapt to diverse network conditions and suffer from issues such as prolonged CDN switching delays and/or complex implementation requirements, resulting in poor QoE. To address these limitations, we present StreamWise---a learning-based solution for real-time CDN selection that works effectively for both on-demand and live adaptive video streaming. StreamWise implements on the content steering standard, leveraging a deep reinforcement learning (DRL) framework to learn and predict in real-time the optimal CDN selection policy by continuously interacting with the environment. Our solution adapts in real-time to network conditions, content characteristics, and viewer requirements ensuring the delivery of high-quality content to users. We evaluate StreamWise through extensive trace-driven emulation-based experiments, demonstrating its superior performance compared to conventional CDN selection strategies. Our results demonstrate a significant improvement in VMAF by ~8.5% with ~1.5x higher average bitrate and QoE improvement of ~48% with minimal rebuffering events for video-on-demand streaming and an improvement in VMAF by ~7%, with ~1.2x higher average bitrate and a QoE improvement of ~22% while substantially reducing the rebuffering events by ~10× for live streaming.

Point cloud video datasets enriched with user behavior traces are critical for advancing tile-based HTTP adaptive streaming (HAS) systems, particularly those reliant on large training data for player modules like Field of View (FoV) prediction. This paper presents MazeLab, a dynamic volumetric video dataset comprising a feature-rich point cloud representation of a large maze environment. The dataset captures navigation traces from 15 participants interacting with 15 distinct maze variants, categorized into seven classes designed to elicit specific behavioral characteristics such as navigation patterns, attention hotspots, and interaction dynamics. By incorporating diverse environmental designs, each targeting unique user responses, MazeLab provides a comprehensive behavioral repository, enabling robust testing and optimization of HAS-driven volumetric video systems in complex scenarios. The dataset is publicly available here.

The recent introduction of the ETSI TS 103 998 content steering standard marks a significant milestone in the evolution of media delivery for content providers. This standard simplifies the process of utilizing multiple Content Delivery Networks (CDNs), a key requirement for managing large-scale client bases. It offers clear guidelines for real-time CDN state switching, which enhances the user experience by dynamically selecting the most suitable CDN based on changing network conditions. In contrast, current industry solutions often resemble load balancing techniques, relying on heuristic approaches that are manually crafted and unable to adapt effectively to varying network environments. These static solutions frequently fail to optimize the user experience in real-time, especially when faced with diverse and unpredictable network conditions. In this paper, we utilize our previous work StreamWise---a DRL-based multi-CDN solution that outperforms existing state-of-the-art methods, offering superior adaptability and efficiency. We integrate this solution into the Dash.js reference player. Through a comprehensive demonstration, we illustrate how StreamWise, or any other multi-CDN solution, can be seamlessly incorporated into Dash.js, providing a benchmark for future developments in multi-CDN switching solutions. Experimental results in the wild demonstrate the performance of various multi-CDN solutions for Dash.js in Video on Demand (VoD) streaming mode. On average, StreamWise provides a ~78% improvement in VMAF compared to other solutions, while also ensuring smooth quality transitions., while also ensuring smooth quality transitions.

QUIC is rapidly emerging as the de facto standard for video streaming, particularly with the advent of 5G technology enabling the delivery of high resolution 4K content. However, sudden fluctuations in network capacity between 4G and 5G networks can occur due to various factors, such as network congestion, user mobility, or physical obstructions, leading to rapid changes in available bandwidth, which pose significant challenges for HTTP Adaptive Streaming (HAS) sessions that rely on more stable data rates for smooth playback, often resulting in interruptions and degraded Quality of Experience (QoE). Furthermore, current QUIC implementations are characterized by static configuration parameters, including constant pacing rates and static buffer sizes. This rigidity limits the server's ability to adapt dynamically to the client's real-time network conditions, making it challenging to prevent playback stalls and improve user QoE. To address these issues, we propose Connection-Aware QUIC (CAQ), which utilizes encrypted client playback statistics to dynamically adjust server parameters and implement an adaptive pacing rate. This approach aims to optimize streaming performance and enhance user experience, even in highly fluctuating network conditions. To evaluate the performance of CAQ, we conduct a series of trace-driven experiments across two key streaming modes: Video-on-Demand (VoD) and Low-Latency Live (LLL). Experimental results demonstrate that CAQ improves the VMAF by ~4%, QoEyin by ~41% while reducing rebuffering duration by ~8% for VoD mode and improves the VMAF by ~7%, QoEyin by ~49% with comparable rebuffering for LLL mode.

QUIC is rapidly emerging as the de-facto standard for HTTP Adaptive Streaming (HAS). QUIC relies on heuristic-based congestion control algorithms which were predominantly designed for TCP and thus have poor generalizability, ultimately degrading the user's Quality of Experience (QoE). Existing learning-based solutions for TCP are not pluggable and require a lot of engineering work, impacting their integration with QUIC. To tackle these challenges, we develop Aero--- the first learning-based plug-and-play congestion control algorithm for QUIC that considers the varying network statistics and can easily be integrated with any QUIC implementation. To analyze the performance of Aero, we conduct a series of comprehensive trace-driven experiments and evaluate its efficiency not only from a congestion control perspective but also the impact it has on the client-driven adaptive bitrate scheme (ABR). Experimental results demonstrate that Aero improves VMAF by ~12% with ~65% less rebuffering for low latency live streaming sessions compared to its competitors. Moreover, Aero excels in terms of delivery rate and delay across different network conditions.

In the realm of 360° video streaming, three paramount challenges arise: accurate prediction of the field of view, efficient bandwidth utilization, and seamless viewport sharing. Current solutions often neglect the latter aspect of 360° videos, thereby necessitating ultra-low latency, high quality, and real-time rendering. The considerable size of 360° videos further intensifies these challenges. To tackle these issues, we introduce LCR360, a deep learning-based neural network solution designed for accurate head movement prediction. Our solution is deployed on E3PO [9], an end-to-end open-source evaluation platform dedicated to 360° video-on-demand streaming. By integrating simulated streams into the Unreal Engine, we provide an immersive experience that enables users to share viewports seamlessly. Moreover, LCR360 surpasses its competitors by ~5% in the S-metric, a measure that combines the MSE of video quality, bandwidth, and storage costs.