3D CNN Micro Expression Recognition Algorithm

This repository contains a 3D Convolutional Neural Network (3D CNN) implementation for micro expression recognition. The algorithm processes video datasets, performs recognition tasks, and offers both GUI and code-based methods for training and testing.

3-Class Composite Dataset

The model is trained and tested on a 3-class composite dataset consisting of the following emotion classes:

• Negative
• Positive
• Surprise

The composite dataset is created by integrating data from the following micro-expression datasets:

1. SAMM (Spontaneous Micro-expression Database)
   A dataset with high-resolution videos of micro-expressions captured under controlled lab conditions.

2. CASMEII (Chinese Academy of Sciences Micro-expression II)
   A well-known dataset containing spontaneous micro-expressions captured under strict illumination and experimental conditions.

3. CAS(ME)²
   A refined dataset for micro-expression analysis with annotations for macro- and micro-expressions.

4. CASME3
   An advanced dataset designed for analyzing micro-expressions with improved diversity and quality of videos.

These datasets provide a diverse set of micro-expressions, enabling the model to generalize effectively across different individuals and scenarios.

Folder Structure

data_processing_gui/

The data_processing_gui folder contains a Graphical User Interface (GUI) for processing video datasets and testing the micro expression recognition algorithm. The GUI allows users to:

• Process and preprocess video files into usable datasets.
• Dynamically select frames for training and evaluation.
• Test the algorithm on processed video datasets with ease.

This GUI simplifies the process of dataset management and testing, making it accessible for users without requiring direct coding.

training/

The training folder contains the files necessary for training the 3D CNN model. It provides two ways to train the model:

1. Using train.py (Local Python Script)

   • The train.py script allows users to train the model on their local machine or server.
   • This method provides full control over the training process and can be customized according to your hardware and dataset.

2. Using Google Colab Notebook

   • Alternatively, you can use the provided Google Colab notebook for training the model in the cloud, eliminating the need for local setup.
   • The Colab notebook offers a cloud-based environment for training, monitoring the model's progress, and evaluating its performance.
   • This is the preferred method for users with limited hardware resources or those who prefer not to manage local environments.

Requirements

Before running the code, ensure that the required dependencies are installed. You can install them by running:

pip install -r requirements.txt