Human Emotion Detection Repository

======================================

This repository is structured to provide a comprehensive overview of our Human Emotion Detection project, including data preparation, model architectures, training details, and performance analysis. Below is a detailed explanation of the repository structure:

Repository Structure

1. data/

Contains all datasets in TFRecord format, organized into three categories:

• raw/: Original dataset without any augmentations.

• augmented/training/: Dataset with standard augmentations applied.

• cut_mix_augmented/training/: Dataset with CutMix augmentation, enhancing data diversity.

2. models/

Stores training history, architectural details, and metrics for each model used in the project:

• EfficientNetB4/: Files related to the EfficientNetB4 architecture, including updates and fine-tuning details.

• EfficientNetB4_finetuned/: Files and performance metrics for the fine-tuned EfficientNetB4 architecture.

• LeNet-5/: Contains files for the LeNet-5 model and its performance metrics.

• ResNet-34/: Includes files and results for the ResNet-34 model.

3. notebooks/

Detailed and well-documented Jupyter notebooks that explain each step of the project and experiments:

• Human_Emotion_Detection_through_Computer_Vision_ComprehensiveGuide.ipynb: A comprehensive guide to the entire project, including methodology and key insights.

• Human_Emotion_Detection_using_EfficientNetB4_fine_tuning.ipynb: Notebook detailing the fine-tuning process for the EfficientNetB4 model.

• Human_Emotion_Detection_using_ResNet_Architecture.ipynb: Details the use of the ResNet architecture for emotion detection.

• Human_Emotion_Detection_using_Transfer_Learning_with_EfficientNetB4.ipynb: Explains the transfer learning approach using EfficientNetB4.

• Exploring VGG16/ EfficientNetB5/EfficientNetB4 for Human Emotion Detection with Grad-CAM Visualization.ipynb : For intermediate layers visualizations , Grad-CAM method , additional training for VGG16 and EfficientNetB5

4. results/

Contains visualizations, metrics, and performance summaries for each model:

• EfficientNetB4/: Visualizations and metrics for the EfficientNetB4 model.

• EfficientNetB4_finetuned/: Results and visualizations for the fine-tuned EfficientNetB4 model.

• LeNet-5/: Performance analysis and metrics for the LeNet-5 model.

• ResNet-34/: Results and visualizations for the ResNet-34 model.

5. utils/images/

Auxiliary folder containing images used throughout the project, such as architecture diagrams and result visualizations.

Releases

Each model version has been packaged as a release, including detailed explanations and usage instructions.

Usage Instructions

1. Navigate to the relevant folder to access data, models, or notebooks.

2. Refer to the notebooks for detailed step-by-step explanations and implementation details.

3. Explore the results/ folder for performance metrics and visualizations.

This structure ensures clarity and ease of navigation for researchers and developers working on the Human Emotion Detection task.