About

The Model

NHSPID, New Hampshire Spider Identifier, is a machine learning tool that identifies spiders in NH.

The Classifier

  • Trained on 123,099 images across 161 species of NH spiders
  • Uses a DINOv2 Vision Transformer (ViT-L/14, 518px) architecture (~300M parameters)
  • Achieves 91.7% accuracy on test data
  • Hosted via Hugging Face Inference Endpoints (Intel Sapphire Rapids, 4 vCPUs, 8 GB)
  • Currently has one head with 161 classes
  • Fine-tuned with differential learning rates (0.0001 for backbone layers, 0.001 for classifier)

The Dataset

The model was trained on observations made available on iNaturalist. The full training dataset is publicly available on Hugging Face.

Inclusion Criteria

  • At least one research-grade observation in New Hampshire
  • A minimum of 30 observations and 75 photos per species
  • Permissively licensed photos only (CC0, CC-BY, CC-BY-SA, CC-BY-NC, CC-BY-NC-SA)

Versions

VersionDateModelSpecies
v17/15/25ViT-L/16 (224)36
v25/28/26DINOv2 ViT-L/14 (518)161

Future Directions

  • Include spiders from bordering regions (e.g., Vermont, Maine, Massachusetts, and southern regions of Canada that border NH)
  • Add species with low image counts
  • Continue to improve the model's accuracy
  • Display default images for both sexes of a species

Credits

Species images from iNaturalist .

Phidippus audax on landing page from wirestock on Freepik .

Contact

Questions, comments, or feedback about the New Hampshire Spider Identifier? Feel free to reach out by email.