Annotated Insect Image Dataset for InsectMorphoAI Software

Description This dataset supports the manuscript "InsectMorphoAI: A Deep Learning Framework for Automated Estimation of Insect Length, Volume, and Biomass". It contains all annotated image data used to train the deep learning models integrated into InsectMorphoAI, along with the validation images and ground-truth biomass measurements used to evaluate the segmentation-based volume estimation pipeline. The dataset is provided to ensure full reproducibility of the research and to serve as a resource for the community working on automated, image-based insect morphometrics. Dataset Components 1. General-Purpose OBB Dataset (Rapid Length Estimation Module)Purpose: Training data for the OBB module, which performs rotation-invariant linear length estimation across diverse insect taxa.Content: 815 images spanning multiple families within Diptera, Hymenoptera, and Coleoptera.Format: Images (.jpg, .png) with corresponding annotations in YOLO OBB format (.txt). 2. Taxon-Specific Segmentation Dataset (Volumetric Analysis Module)Purpose: Training data for the instance segmentation module, which delineates head, thorax, and abdomen to extract curvilinear length and approximate 3D body volume. The model distributed with the software is specialized for Tachinidae (Diptera).Content: 1,320 images of representative tachinid species, structured across two annotation stages: initial_labeling_strategy_820_images: Initial annotations covering only directly visible body parts. final_refined_dataset_500_images: Refined annotations explicitly including inferred outlines for body parts partially obscured by wings or legs, used to train the final production model. Format: Images (.png) with corresponding annotations in YOLO segmentation format (.txt). 3. Biomass Validation DatasetPurpose: Independent validation of the volume-to-biomass pipeline. These images were processed by the trained segmentation model to generate volume estimates, which were then correlated against physical weight measurements.Content: 100 tachinid specimens representing 23 species (mean body length: 9.36 mm; mean body-only dry weight: 6.1 mg). Includes both wet weight and body-only dry weight (BDW) measurements obtained using calibrated semi-micro scales. Ground-truth measurements are provided in the accompanying Supplementary Data file (Validation_Measurements sheet).Format: Images (.png, .jpg). General Information Image Acquisition: All images were captured using the Entomoscope (Wührl et al., 2022), a high-resolution macro-photography system. Multiple focal planes were combined using Helicon Focus to achieve full depth-of-field clarity. All specimens were preserved and imaged in ethanol. Specimen Metadata: Complete specimen information, image identifiers, and ground-truth validation measurements are provided in the Supplementary Data file accompanying the publication (Tables S1–S2 and Validation_Measurements sheet). Reproducibility: Model weights, training code, and inference software are distributed separately via the InsectMorphoAI GitHub repository and Docker image.