Split definitions and corpus statistics for "Bench Characterization of Lightweight Object-Detection Models on an Edge-AI Camera for UAV-Oriented Source-Water Monitoring

This deposit accompanies the manuscript "Bench Characterization of Lightweight Object-Detection Models on an Edge-AI Camera for UAV-Oriented Source-Water Monitoring". The imagery and annotations used in the study are obtained from AI-Hub and cannot be redistributed under the AI-Hub data-use terms. This deposit therefore contains no imagery and no annotation geometry. It contains the exact file lists of the training and validation splits and the class-balance and acquisition statistics computed from them, so that a reader who obtains AI-Hub access can reconstruct the identical splits and verify the reported composition. Contents: - train_files.txt, val_files.txt: the file names forming each split (4,813 and 575 images). - dataset_statistics.json: class balance, object-size distribution, acquisition dates and administrative regions, and the session-level relationship between the two splits. The source corpora are the AI-Hub River-Influent Agricultural and Livestock Non-Point Source Pollution Image Data and the AI-Hub Coastal Pollutant Image Data, available from https://aihub.or.kr subject to AI-Hub access approval.