Audio Classification of the Content of Food Containers and Drinking glasses (Pre-trained TensorFlow models)

This upload contains both neural network's architecture and pre-trained weights, used in the study Audio Classification of the Content of Food Containers and Drinking Glasses. The weights are trained on the training split of the CORSMAL Containers Manipulation dataset, under the setup described in the paper.

More information can be found at the project webpage.

acc_pretrained_models.zip contains the pre-trained models of the proposed ACC network. In order to load the pre-trained models, the files listed below should be moved into the project directory:

/acc-net/methods/acc/models/

• acc_action.json
• acc_action.h5
• acc_pouring.json
• acc_pouring.h5
• acc_shaking.json
• acc_shaking.h5

Note: you might need to create the last directory (models).

These models can be either our pre-trained models or your own exported models after training.

models_comparison.zip contains the pre-trained models of the methods under comparison for the validation section of the paper:

• Deep-Learning baselines (TensorFlow models):
  • VGG-11
  • ResNet-18
  • ImageNet ResNet-18 (Transfer Learning)
  • ResNet-14
• Machine Learning baselines (Scikit-learn models):
  • k-Nearest-Neighbors
  • Support Vector Machine
  • Random Forest

These methods are also trained on the training split of the CORSMAL Containers Manipulation dataset