Dafisns/human-emotion-recognition

Human Emotion Recognition

Deep learning models for classifying human facial emotions.

Emotion Classes

• 😠 Angry
• 😨 Fear
• 😊 Happy
• 😢 Sad
• 😲 Surprise

Model Performance

Model	Test Accuracy	Test Loss	Epochs
Base CNN	92.41%	0.268	33
MobileNetV3Small	81.56%	0.551	50

🏆 Best model: Base CNN with 92.41% test accuracy

Models

File	Format	Input Size	Description
model_base.h5	Keras H5	128x128x1	Custom CNN (Grayscale)
model_transfer_learning.keras	Keras	224x224x3	MobileNetV3Small (RGB)
tflite/best_model.tflite	TFLite	128x128x1	Mobile/Edge
tfjs_model/	TF.js	128x128x1	Web deployment

Usage

Python

from huggingface_hub import hf_hub_download
import tensorflow as tf
import numpy as np

# Download model
model_path = hf_hub_download(
    repo_id="dafisnadhif/human-emotion-recognition",
    filename="model_base.h5"
)

# Load model
model = tf.keras.models.load_model(model_path)

# Predict
CLASS_NAMES = ['Angry', 'Fear', 'Happy', 'Sad', 'Surprise']
predictions = model.predict(img_batch)
print(CLASS_NAMES[np.argmax(predictions[0])])

TensorFlow Lite

from huggingface_hub import hf_hub_download
import tensorflow as tf

tflite_path = hf_hub_download(
    repo_id="dafisnadhif/human-emotion-recognition",
    filename="tflite/best_model.tflite"
)

interpreter = tf.lite.Interpreter(model_path=tflite_path)
interpreter.allocate_tensors()

TensorFlow.js

const model = await tf.loadLayersModel(
  'https://huggingface.co/dafisnadhif/human-emotion-recognition/resolve/main/tfjs_model/model.json'
);

Training Details

Parameter	Value
Dataset	Human Face Emotions
Images	~47,000 facial images
Source Code	GitHub
Framework	TensorFlow 2.x / Keras
Optimizer	AdamW (lr=1e-3, weight_decay=1e-4)
Loss	Sparse Categorical Crossentropy
Batch Size	256
Callbacks	EarlyStopping (patience=8), ReduceLROnPlateau

License

MIT License

Author

Dafis Nadhif Saputra