EmotionRecognition

The simplest TensorFlow Emotion Recognition implementation using TensorFlow.

More details in Medium article: https://aruno14.medium.com/very-simple-emotion-recognition-290bd1db234e

Dataset

• FER2013 Dataset: https://www.kaggle.com/msambare/fer2013

Folder structure

• /
• /train_emotion.py
• /predict_emotion.py
• /emotions/[test, train]/[angry, disgust, fear, happy, neutral, sad, surprise]

Model

• Input size: (48, 48, 1)
• Accuracy: 0.5658

How to use

1. Download FER2013 dataset
2. Create emotions folder
3. Extract train and test folder in emotions folder
4. Run train_emotion.py
5. Add a face file named test.jpg
6. Run predict_emotion.py

Convert to TensorFlowLite

python3 convert2tflite.py --model_folder=model_emotion/ --output=model_emotion.tflite

Convert to TensorFlowJS (with quantization)

tensorflowjs_converter model_emotion/ quantized_model_emotions/ --input_format tf_saved_model --output_format tfjs_graph_model --quantize_float16