Automated Dataset Profiling and Exploratory Validation in the wyoloservice2 Stack

The wyoloservice2 worker library includes a pre-train Exploratory Data Analysis (EDA) module that profiles YOLO datasets for class balance, bounding-box geometry, and image quality. We document this component, its implementation, standalone benchmark results, and a comparative evaluation against established data validation tools. The EDA module detects corrupt images, duplicate files, class imbalance, and bbox anomalies in 5.2 to 19.1 seconds for datasets up to 10k images, scaling to 142 seconds for 100k images on commodity CPU. We benchmark our EDA against Great Expectations, Deepchecks, Cleanlab, DVC metrics, Evidently AI, and WhyLabs across 50 empirical computer vision datasets.