Annotated Insect Image Dataset for InsectMorphoAI Software

Description This dataset supports the manuscript, "InsectMorphoAI: A deep learning-based software for rapid and non-invasive estimation of insect length, volume, and biomass,". It contains all the annotated image data used to train and validate the deep learning models integrated into the InsectMorphoAI software package. The dataset is provided to ensure the full reproducibility of our research and to serve as a valuable resource for the community working on automated, image-based analysis of insects. Dataset Components The dataset is divided into two primary components, corresponding to the two analysis modules in the InsectMorphoAI software: 1. General-Purpose Dataset (for "Rapid Scan" / OBB) Purpose: Used to train the general-purpose model for the "Rapid Scan" module, which performs quick linear length estimation. Content: Contains 815 images of insects from diverse families within the orders Diptera, Hymenoptera, and Coleoptera. Format: Images (.jpg, .png) and corresponding annotations in YOLO OBB format (.txt). 2. Taxon-Specific Dataset (for "Detailed Analysis" / Segmentation) Purpose: Used to train the high-precision model for the "Detailed Analysis" module, which performs curvilinear length, volume, and biomass estimation. The model included with the software is specialized for Tachinidae (Diptera). Content: Contains 1,320 images of various representative tachinid species. Annotation Strategy: A two-stage annotation strategy was employed, and the dataset is structured to reflect this: initial_labeling_strategy_820_images: Data corresponding to the initial annotation approach, focusing only on directly visible body parts final_refined_dataset_500_images: This is the dataset used for the final model. The key refinement in this stage was the explicit annotation of inferred outlines for body parts partially obscured by wings or legs, where reliable estimation was possible. Format: Images (.png) and corresponding annotations in YOLO segmentation format (.txt). General Information Image Acquisition: All images were captured using a high-resolution DIY microscope, the Entomoscope (Wührl et al., 2024). Multiple focal planes were stacked using Helicon Focus to achieve optimal clarity. All specimens were preserved and imaged in ethanol. Image Diversity: For both datasets, images were captured from various views and orientations to ensure the models are robust to differences in specimen positioning. Metadata: Detailed specimen information and the ground-truth validation measurements used in the manuscript are provided in the Supplementary Data file accompanying the publication.