MOSMON-Larvae

This record contains the main MOSMON-Larvae dataset release, including annotated mosquito-larvae images, COCO and YOLO annotation formats, extracted video frames, original still photographs, metadata files, and processing scripts. MOSMON-Larvae is a multispecies mosquito-larvae dataset developed to support computer-vision research on larval detection, tracking, motion analysis, and automated vector surveillance. The dataset was collected under controlled indoor laboratory conditions while preserving naturalistic aquatic variability, including changes in illumination, water turbidity, reflections, larval density, container configuration, organic debris, and camera viewpoint. The dataset includes mosquito larvae from four medically relevant species: Aedes albopictus, Aedes aegypti, Culex pipiens, and Anopheles stephensi. Larvae were primarily recorded at developmental stages 3–4. Species identity and acquisition conditions are encoded in the structured filenames and associated metadata. This record provides the image-based and metadata components of the MOSMON-Larvae dataset. The release includes high-resolution annotated images, bounding-box annotations in COCO and YOLO formats, extracted frames from source videos, original still photographs, dataset summary statistics, documentation, and scripts used for metadata extraction, validation, and analysis. The files are distributed as ZIP archives to simplify download and preserve the internal folder structure of the dataset. This record contains 87 uploaded files: 69 extracted-frame ZIP archives: all_extracted_frames_001.zip to all_extracted_frames_069.zip 11 original-photo ZIP archives: all_photos_001.zip to all_photos_011.zip Main dataset and metadata archives/files: mosmon_larvae_coco.zip: annotated image dataset in COCO format mosmon_larvae_yolo.zip: annotated image dataset in YOLO format scripts.zip: Python utilities for dataset statistics, metadata extraction, validation, and figure generation summary.json: aggregated dataset statistics and metadata summary readme.txt: dataset organization and usage documentation filename_mapping.tsv: mapping between shortened Zenodo filenames and original archive paths checksums.sha256: SHA256 checksums for file-integrity verification The original filenames contain structured acquisition metadata, including recording date, project identifier, species, larval stage, camera model, acquisition modality, resolution and frame rate when applicable, lens or distortion configuration, lighting condition, camera position, container type, and water depth. Users should consult filename_mapping.tsv to recover the original archive paths and associated acquisition context for each shortened Zenodo filename. The annotated images are provided in both COCO and YOLO formats. The COCO release includes JSON annotation files with bounding boxes encoded in absolute pixel coordinates. The YOLO release includes one text annotation file per image, with normalized bounding-box coordinates. Both annotation formats derive from the same curated bounding-box dataset and are intended to support compatibility with common object-detection frameworks. The extracted-frame archives contain frame sequences sampled from the original source videos. Frames are grouped by source-video identity to preserve traceability between extracted images and raw footage. Not all extracted frames are annotated; annotation proceeded until the target dataset scale was reached. Users performing temporal or video-based experiments should account for temporal correlation between frames and should use video-disjoint splits when evaluating generalization. The photo archives contain original still photographs acquired during the MOSMON-Larvae acquisition campaign. These images preserve the original acquisition context and can support additional annotation campaigns, domain-adaptation studies, and dataset extension. Recommended use - These files are suitable for: mosquito-larvae detection in high-resolution images; small-object detection benchmarking; training and evaluation of object-detection models using COCO or YOLO formats; frame-based analysis of mosquito-larvae videos; domain-adaptation studies across cameras, lighting conditions, and acquisition configurations; dataset validation, statistics reproduction, and metadata inspection; extension of the MOSMON-Larvae dataset through additional annotation or processing. For machine-learning experiments, users should account for the dense small-object nature of the dataset, class imbalance, high image resolution, and possible temporal correlation among frames extracted from the same source video. For video-derived data, video-disjoint splits are recommended to reduce leakage between temporally related samples. File integrity - Downloaded files can be verified with: sha256sum -c checksums.sha256 Related dataset - This record is associated with the broader MOSMON-Larvae dataset and its raw source-video records. Raw videos are released separately by acquisition hardware to provide reproducible access to the original video material used to derive extracted frames. Please see also: G. Di Lorenzo et al., “MOSMON-Larvae: Raw Insta360 Ace Pro 2 Videos of Mosquito Larvae”. Zenodo, Jun. 22, 2026. doi: 10.5281/zenodo.20803118. G. Di Lorenzo et al., “MOSMON-Larvae: Raw DJI Osmo Action 5 Videos of Mosquito Larvae”. Zenodo, Jun. 23, 2026. doi: 10.5281/zenodo.20809111. G. Di Lorenzo et al., “MOSMON-Larvae: Raw GoPro Hero 11 Videos of Mosquito Larvae”. Zenodo, Jun. 23, 2026. doi: 10.5281/zenodo.20815841. G. DI Lorenzo et al., “MOSMON-Larvae: Raw GoPro Hero 13 Videos of Mosquito Larvae”. Zenodo, Jun. 23, 2026. doi: 10.5281/zenodo.20819747. G. Di Lorenzo et al., “MOSMON-Larvae”. Zenodo, Jun. 24, 2026. doi: 10.5281/zenodo.20821176. Citation - If you use this dataset, please cite this Zenodo record and the associated MOSMON-Larvae dataset or paper when available.