Birdbrain

About

Birdbrain is a fine-grained bird species classifier with models trained on CUB-200-2011 and iNaturalist 2021 (birds only).

What it does

Given a bird photo, a model predicts a species and returns a confidence score plus the top five candidates. See the full supported species list. Training uses transfer learning with ImageNet-pretrained EfficientNet-B0 backbones and staged fine-tuning.

Models

  • BirdBrain Voyager — global bird specialist trained on the combined CUB + iNat species set; default on the Identify page
  • EfficientNet-B0 (CUB) — 200 North American species
  • ResNet-50 (CUB) — same 200-species CUB label set with a deeper backbone

Project structure

  • training/ — YAML-driven training, evaluation, and analysis
  • web/ — this SvelteKit frontend
  • api/ — FastAPI inference service
  • docs/ — technical documentation for the ML pipeline

Deployment

Live at birdbrain.djm-apps.com. The site calls POST /api/predict; training and serving stay separate processes in one monorepo.