Emotion Recognition Model (MobileNetV2 + FER2013)

This model classifies facial emotions into 7 categories: angry, disgust, fear, happy, neutral, sad, surprise. It is based on transfer learning using MobileNetV2 and fine-tuned on the FER2013 dataset.

🧠 Training Summary

Architecture: MobileNetV2 + custom classification head
Dataset: FER2013
Training: 15 epochs + 10 fine-tuning
Accuracy: 0.52
Weighted F1-score: 0.48

📈 Performance

The model performs best on: happy, surprise, neutral. Challenging classes: disgust, fear.

Confusion matrix and accuracy/loss curves are included.

🚀 Usage

from tensorflow.keras.models import load_model
model = load_model("final_emotion_model_finetuned.h5")

🖼️ Classes

0 = angry
1 = disgust
2 = fear
3 = happy
4 = neutral
5 = sad
6 = surprise

🔐 License

Apache 2.0

Downloads last month: 6

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support