Face Emotion Recognition CNN (v2)
Model Description
Face Emotion Recognition CNN (v2) is a lightweight Convolutional Neural Network designed to classify facial expressions into seven basic emotion categories using grayscale facial images of size 48×48. The model is optimized for fast inference and practical deployment in real-time and resource-constrained environments.
The model was trained using the Adam optimizer and categorical cross-entropy loss, balancing performance and computational efficiency.
Intended Use
This model is intended for:
- Facial emotion recognition from static images
- Human–computer interaction systems
- Emotion-aware applications
- Academic research and educational projects
Out-of-Scope Use
The model should not be used for:
- Surveillance, biometric identification, or user profiling
- Medical, psychological, or legal decision-making
- Any high-risk or safety-critical applications
Model Architecture
- Architecture Type: Convolutional Neural Network (CNN)
- Input: Grayscale facial images (48 × 48)
- Output: Probability distribution over 7 emotion classes
- Optimizer: Adam
- Loss Function: Categorical Crossentropy
Training Configuration
- Image Size: 48 × 48
- Batch Size: 64
- Epochs: 40
- Learning Rate: 0.0005
- Number of Classes: 7
- Model Name: face_emotion_cnn_v2
Emotion Labels
| Label ID | Emotion |
|---|
| 0 | Angry |
| 1 | Disgust |
| 2 | Fear |
| 3 | Happy |
| 4 | Neutral |
| 5 | Sad |
| 6 | Surprise |
Dataset
The model was trained on a facial emotion dataset containing grayscale face images resized to 48×48 pixels. Images were preprocessed using normalization and face alignment where applicable. The dataset includes seven emotion categories.
Note: Dataset composition and demographic distribution may influence model bias and generalization.
Evaluation
The model was evaluated on a held-out test set with the following results:
Training Performance (Final Epoch)
- Accuracy: 0.5976
- Loss: 1.1217
Test Performance
These results reflect moderate performance, which is typical for compact CNN models trained on low-resolution facial emotion datasets.
Usage
Inference Example
1from transformers import pipeline
2
3classifier = pipeline(
4 task="image-classification",
5 model="your-username/face_emotion_cnn_v2"
6)
7
8predictions = classifier("face.jpg")
9print(predictions)
