/emotion-theme-recognition-app

Music emotions and themes classifier app could recognize 56 classes using three trained models (based on ResNet50, InceptionNetV2, EfficientNetB3), applying the transfer learning approach.

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

Emotion-theme-recognition-app

An interactive web application which could recognize music emotion and themes using one of three pretrained models and return the probability in percentages.

The project was influenced by the task "Emotions and Themes in Music" of MediaEval.

The first steps

  • Clone this repo
  • Install requirements
  • Download the three directories with pretrained models from this Google Drive link and place them all to the models directory.
  • Run the script passing appropriate arguments

👇 Screenshot:

Models

You could choose which model you want to use at first. There are three accessable variants of models: ResNet50, InceptionNetV2, EfficientNetB3. Implementations of preprocessing stage is located in repository Emotion-and-Theme-Recognition-in-Music-Task and training the model stage - in *.py files with hold the names of the available models, located in current repository.