A cross-platform desktop app that uses deep learning to recognize emotions from facial images or video using 8 models (TensorFlow + PyTorch). Developed with React + Electron + FastAPI.
- Frontend: ReactJS, Electron, FaceAPI.js
- Backend: Python, FastAPI
- ML: TensorFlow, PyTorch (VGG16 models)
- Real-time webcam-based emotion recognition
- Upload image/video for analysis
- Compare 8 emotion models
- Fully local app (no cloud required)
- FER2013
- KDEF
- NHFI
- Mixed Dataset
Each dataset was used to train two models (TensorFlow + PyTorch). A two-phase transfer learning strategy was used with VGG16 as the base model.
| Dataset | Framework | Accuracy | F1 Score |
|---|---|---|---|
| KDEF | TensorFlow | 92% | 0.92 |
| KDEF | PyTorch | 91% | 0.91 |
| NHFI | TensorFlow | 78% | 0.76 |
| NHFI | PyTorch | 72% | 0.71 |
| FER2013 | TensorFlow | 67% | 0.65 |
| FER2013 | PyTorch | 66% | 0.63 |
| Mixed | TensorFlow | 70% | 0.70 |
| Mixed | PyTorch | 68% | 0.68 |
cd Backend
pip install -r requirements.txt
uvicorn app.main:app --reload