Annotated In-Trawl Camera Images of Mesopelagic Organisms: A Dataset for Ecological and Machine Learning Applications

This repository provides a structured collection of datasets and configuration files used for training and evaluating object detection models to detect mesopelagic organisms from images.

The datasets (described in the readme file) are stored in CSV format under datasets/csv_files, where each file contains image paths, bounding box coordinates, species class labels, and image names. Multiple dataset variants (described in the manuscript) are
included, covering different colour corrections (e.g., red CC, red non-CC, white), augmentation strategies, and splits (train, validation, test). For YOLO-based training, the data is organized under datasets/YOLO, with separate folders for image paths (images/),
YOLO-format labels (labels/), and dataset configuration files (.yaml).

The deepvision/ directory contains annotated data, separated into rectified and non-rectified formats, along with original XML annotations and session-level metadata in meta_per_deep_vision_session.csv as well. It also contains synthetic data used in some dataset
configurations for training, under /train.

The model/ directory contains an example training script and a configuration file with the parameters used for training.

This structure supports flexible experimentation with various data configurations and model training pipelines, facilitating reproducibility and scalability in object detection tasks.