Face Emotion Detection (FastAPI)

日本語

Welcome to my Face Emotion Detection project! This application is designed to detect emotions in faces from uploaded images using a combination of Haar Cascade for face detection and a ViT (Vision Transformer) model for emotion prediction. The model has been trained on 5 emotion classes: Happy, Sad, Angry, Surprised, and Neutral.

Website Preview

Home Page

Test Image(emotion: Happy)
Result page

Usage

1. Upload an image containing a face.
2. The Haar Cascade algorithm will detect the face.
3. The image, converted to a numpy array, is then passed to the ViT Transformer model.
4. The model predicts the emotion of the detected face.

Model Details

• I use a ViT Transformer model pretrained on Google's Patch 16 dataset with an image size of 224x224 pixels.
• The model is fine-tuned on a custom emotion dataset, which includes Kaggle's dataset and additional custom data, to accurately classify emotions.

Adding Custom Dataset and Kaggle

To enhance model accuracy, I added a custom dataset that includes a combination of Kaggle's dataset and additional custom data. This expanded dataset allowed my model to better understand and classify a wider range of emotions. A val Accuracy of 90% was achieved.

Deployment

This project is hosted on Render, which may impact loading times and stability. Due to the nature of hosting service, occasional crashes or slower response times may occur.

Technologies Used

• FastAPI: For building the web application.
• Haar Cascade: For face detection.
• Vision Transformer (ViT): For emotion prediction.
• Google's Patch 16 Dataset: For pretraining the ViT model.
• Kaggle Dataset: Incorporated for comprehensive training.

Running the Application Locally

To run this application locally, follow these steps:

1. Clone this repository.
2. Install the necessary dependencies using pip install -r requirements.txt.
3. Run the FastAPI app using uvicorn app:app --reload.

Notebook Repository

Explore the project's development and model training in detail by visiting my notebook repository here.

Contributing

Contributions are welcome! If you have any suggestions or improvements, please open an issue or submit a pull request.

License

This project is licensed under the MIT License - see the License file for details.

Enjoy exploring and using the Face Emotion Detection application!