SPARK Experimental Dataset: Edge-Assisted XR Pose Estimation

This dataset contains the experimental data collected to evaluate SPARK (Task-oriented Edge-Assisted Pose Estimation for Real-Time Kinematics AR),an edge-assisted architecture for real-time augmented reality (AR) pose estimation on resource-constrained XR devices. The dataset consists of 32 CSV files containing runtime telemetry collected during experiments with different image resolutions, JPEG compressionqualities, marker distances, and motion conditions. The experiments evaluate the impact of visual data compression and downsampling on communicationefficiency and 6-DoF pose-estimation performance. The recorded metrics include frame rate (FPS), received data rate, pose estimation time, normalized pose-estimation accuracy, reprojection/pixelerror, stability, and tracking accuracy. Four image resolutions are evaluated: 1920×1080, 1280×720, 640×360, and 360×240. JPEG quality levels of 100%, 75%, 50%, and 25% are included in thecompression experiments. The dataset also contains experiments performed at different marker distances and under moving-marker conditions. The data support the experimental analysis presented in the associated manuscript, including the evaluation of the trade-off between communicationpayload, bandwidth, latency, and pose-estimation accuracy in an edge-assisted XR system. The dataset contains experimental CSV telemetry only and does not include raw camera or video recordings. This dataset is associated with the manuscript: "SPARK: Task-oriented Edge-Assisted Pose Estimation for Real-Time Kinematics Augmented Reality", submitted to Computer Networks.