Models for MED and MSC

HumBugDB models

This release distributes the binaries for the Bayesian neural network models, to preserve the small size of the repository.

We include models trained for two tasks. The model objects and tasks are described at:

1. MED: Mosquito Event detection
2. MSC: Mosquito Species classification

The models correspond to all of the results reported in the paper HumBugDB: A Large-scale Acoustic Mosquito Dataset.

PyTorch and Keras models

This release distributes the binaries for the Bayesian neural network models, to preserve the small size of the repository. We include five trained models, to be placed in /outputs/models/keras/ or /outputs/models/pytorch/:

• PyTorch BNN-ResNet-50 (paper version): neurips_2021_humbugdb_resnet50_bnn.pth
• BNN-ResNet-18 (paper version): neurips_2021_humbugdb_resnet18_bnn.pth
• Keras BNN-4conv (paper version): neurips_2021_humbugdb_keras_bnn.hdf5
• Keras BNN-4conv (best version): neurips_2021_humbugdb_keras_bnn_best.hdf5
• Keras BNN-4conv (for appendix Test B): neurips_2021_humbugdb_supplement_TestB_keras_bnn.hdf5