ArthroNat

YOLO11 model trained on arthropods over natural backgrounds.

Intended use. ArthroNat is for arthropods photographed in the field, against vegetation, soil and other natural substrate, the setting where an insect is a small part of a visually busy frame. It is the counterpart to FlatBug, which is built for uniform backgrounds.

Architecture and training

• Architecture: YOLO11, as n and l variants
• Task: Object Detection (bounding boxes)
• Trained on: ArthroNat; two training mixes are published: the base ArthroNat data, and a variant additionally trained on FlatBug data
• Taxonomic coverage: not published as a fixed class list in this registry
• Framework: ultralytics

Inputs and outputs

• Input size: 640x640
• Channel order: RGB
• Outputs: bounding boxes
• Input modality: image only (discriminative model)

Performance

Check the source publication for metrics.

Limitations

License and rights

• License: MIT
• Commercial use: allowed
• Restrictions: None stated.
• Upstream source: https://huggingface.co/edgaremy/arthropod-detector

Citation

@article{arthronat2026,
  title   = {ArthroNat: arthropod detection over natural backgrounds},
  author  = {Remy, Edgar et al.},
  journal = {bioRxiv},
  year    = {2026},
  note    = {https://www.biorxiv.org/content/10.64898/2026.05.06.723207v1.full}
}

How to obtain the weights

From Hugging Face, pinned to commit 54edf736. The repository holds eight files, four trained variants, each published in both PyTorch and ONNX form:

No authentication is required.

from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="edgaremy/arthropod-detector",
    filename="yolo11n_ArthroNat_mosaic33.pt",
    revision="54edf7364582e9efda68c04edfd672cfe109fb4a",
)

Output format probe

The author-documented Ultralytics Python interface was probed on CPU with the pinned YOLO11n mosaic33 checkpoint. It returned Results objects with pixel-coordinate boxes; no JSON export was requested. Other weight variants were not run.

Probe date: 2026-09-30. Reproducible setup and evidence are in src/probe.