MAMBO V3 — PyTorch and ONNX deployment

MAMBO V3 classifies moths and butterflies at species, genus and family level,
covering 12,632 species with shared PyTorch and ONNX interfaces.

• Standalone ONNX deployment without PyTorch or a GPU.
• Automatic model downloads, verified caching and offline use.
• Regional presets, including the legacy V2 European lists, and custom class lists.
• Optional test-time augmentation, embeddings and streaming inference.
• V2/V3 quality comparisons on monitoring images and in-domain photographs,
  alongside CPU, laptop GPU and B200 throughput measurements.

Quick start

uv pip install 'mambo-v3[onnx]==0.3.0'
mambo_predict -i moth.jpg -o output --name predictions

Defaults are ONNX, CPU and the global species list.

Integration guide and comparisons
Model card and weights
Interactive demo

Weights: CC BY-NC-SA 4.0 (non-commercial, share-alike).
Deployment code: MIT.

See the integration guide for V2 migration, configuration and compatibility details.