How houdaipha/morad works
Project Type & Purpose
• This is a multimodal emotion recognition framework that benchmarks various models (unimodal and multimodal) on a curated dataset (MORAD).
• The core purpose is to evaluate deep learning models across different modalities, as evidenced by the diverse configurations and modules for audio, vision, text, and their combinations.
Analyzing the File Structure
• Top-level directories include:
• configs: Contains configuration YAML files for different model setups and modalities. They are grouped by modality (audio, text, vision) and multi (for multimodal configurations).
• data: Provides data processing, including utilities and loaders for different types of input (audio, vision, text, and a unified multi-modal component).
• engines: Implements the core execution logic for each model during training or evaluation. Modules are again separated by modality.
• models: Defines the model architectures for different modalities. They mirror the engines’ directory structure.
• training: Contains scripts for handling the training routines across modalities.
• Additional files like main.py and test.py act as entry points for training and testing respectively.
• The separation of directories by modality (audio, vision, text) and by purpose (data, engines, models) clearly indicates a modular design that can be diagrammed into layers.