Large, clean image datasets do not exist for most specialized domains. I build the pipelines that create them, with iterative Grounding DINO detection, SAM and Mask2Former segmentation and VLM completeness checks, and release the resulting datasets with validated measurements.
Over 13,200 NEON ground beetles from 30 sites across the continental US and Hawaii, imaged at high resolution with 39,064 elytra measurements. Measurements from images agree closely with calipers on physical specimens.
An end-to-end workflow from photographed collection trays to curated, AI-ready specimen data: iterative open-vocabulary detection with VLM completeness checks, scale-bar and color-card localization, automated metadata validation, and curation errors sent back to the biorepository.
Three stages for tray images of pinned beetles: detect every specimen, crop and sort it with its metadata, and segment it into body parts. Detection matches the exact count on 97.81% of 1,506 NEON trays.
A practical framework, written by taxonomists, collection managers, ecologists and computer scientists together, for imaging specimens so that computer vision can use them: metadata, positioning, calibration, background, lighting, resolution and file formats.
Pinned carabids from NEON's Hawaii site: 162 group images, 1,614 individual specimen crops and 1,579 trait rows across 14 endemic species, with caliper and image-based elytra measurements.
Ethanol-preserved carabids from 30 NEON sites: 577 group images, 11,654 individual specimen images and 39,064 elytra measurement records, plus segmentation train and test splits.