Emotion Detection from Facial Images

This repository contains the code for an emotion detection model using Convolutional Neural Networks (CNNs) in TensorFlow/Keras. The model is trained and evaluated on a dataset of facial images labeled with emotions (e.g., happy, sad, angry, neutral).

Setup

1. Clone the Repository:

   git clone https://github.com/AngadiRanjeeta/Emotion-Detection.git
   cd Emotion-Detection

2. Open the Colab Notebook:

   • Open emotion_detection.ipynb in Google Colab.

3. Install Dependencies:

   • Install the required dependencies by running:

     !pip install tensorflow numpy pandas matplotlib

4. Upload Dataset:

   • Download the dataset.
   • Upload the downloaded dataset to your Google Drive and mount it in the Colab notebook as shown in the code.

5. Run the Notebook:

   • Execute the cells in the notebook sequentially.

Results

The model achieves the following performance:

• Validation Accuracy: 0.5634
• Test Accuracy: 0.4947

Visualizations

Training and validation accuracy and loss curves are provided to visualize the training process.

Dependencies

• TensorFlow
• Keras
• Matplotlib
• Numpy
• Sklearn
• Seaborn

Approach and Methodology

1. Data Preparation:

   • Defined dataset paths for training, validation, and test sets.
   • Utilized ImageDataGenerator for data augmentation (horizontal flipping) and normalization.

2. Model Architecture:

   • Built a Convolutional Neural Network (CNN) with three convolutional layers followed by max-pooling and dropout layers to prevent overfitting.
   • Used a fully connected layer with dropout before the final softmax layer to classify emotions into 7 categories.

3. Model Compilation:

   • Compiled the model using the Adam optimizer and categorical cross-entropy loss, with accuracy as the evaluation metric.

4. Training:

   • Set up callbacks: ReduceLROnPlateau to adjust the learning rate, ModelCheckpoint to save the best model, and EarlyStopping to stop training if the model's performance didn't improve for a specified number of epochs.
   • Trained the model on the training data while validating it on the validation set.

5. Evaluation:

   • Evaluated the model's performance on both the validation and test sets to calculate accuracy.
   • Plotted the training history to visualize accuracy and loss over epochs.

6. Predictions and Analysis:

   • Predicted emotions on the test set and calculated the accuracy.
   • Generated a confusion matrix and classification report to evaluate performance across all classes.
   • Displayed some misclassified examples for further analysis.

License

This project is licensed under the MIT License.