Spaces:
Sleeping
Apply for a GPU community grant: Academic project
COLLEMBOT — Open-source AI counting tool for OECD 232 ecotoxicology tests
COLLEMBOT is a free, open-source computer vision pipeline for automated counting and body-size measurement of Folsomia candida (Collembola/springtails) in OECD 232 soil reproduction toxicity tests. It is developed by researchers at Eawag (Swiss Federal Institute of Aquatic Science and Technology), FHNW, University of Zurich, Vrije Universiteit Amsterdam, University of Bayreuth, and Cloverstrategy Lda (Portugal).
The tool is published in Environmental Toxicology and Chemistry (2026, DOI: 10.1093/etojnl/vgag068). Model weights are archived on Zenodo (DOI: 10.5281/zenodo.17987886, AGPL-3.0). The training dataset is CC-BY-4.0. The full source code is on GitHub (waldstrom/collembot).
COLLEMBOT is designed to democratise access to automated ecotoxicology analysis. Manual Collembola counting is time-consuming, error-prone, and a significant bottleneck for labs with limited personnel. By hosting COLLEMBOT on HuggingFace with GPU support, any researcher with a browser and an image can run a full OECD 232 analysis in under a minute — with no installation, no hardware investment, and no cost. This is particularly important for research groups in low- and middle-income countries where access to computational infrastructure remains a major barrier to conducting standardized ecotoxicology testing.
The tool includes automated organism counting, body-size measurement, multi-image batch processing with statistical quality flagging, and a hand-correction interface with full audit trail. GPU access would reduce per-image inference from ~10 minutes (CPU) to ~30 seconds — the difference between a tool that is theoretically accessible and one that is genuinely usable in a daily lab workflow. The tool is currently in active development as a beta release, and community feedback will directly shape future improvements.