About

Birdbrain is a fine-grained bird species classifier trained on the Caltech-UCSD Birds-200-2011 (CUB-200) dataset.

What it does

Given a bird photo, the model predicts one of 200 North American species and returns a confidence score plus the top five candidates. See the full supported species list. Training uses transfer learning with ImageNet-pretrained backbones (EfficientNet-B0 and ResNet50) and a five-stage fine-tuning pipeline.

Models

EfficientNet-B0 — staged checkpoints under birdbrain_v1*.pt
ResNet50 — staged checkpoints under birdbrain_resnet50_v1*.pt

Production deployment will serve a single best checkpoint with matching preprocessing (image size, normalization, optional bounding-box crop).

Project structure

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

Deployment

Planned home: birdbrain.djm-apps.com. The site will call a backend POST /api/predict endpoint; training and serving stay separate processes in one monorepo.