📱 AgeSense

A lightweight, mobile-friendly age prediction model optimized for deployment across multiple platforms (PyTorch, ONNX, CoreML, TFLite).

---

🎯 Overview

AgeSense is a mobile-optimized deep learning model for predicting age from facial images. The model leverages a modified MobileNetV2 architecture that benefits from enhanced data augmentation and extended fine-tuning for improved performance. It is provided in multiple formats (PyTorch, ONNX, CoreML) so you can easily integrate it into iOS, Android, or web applications.

---

🔍 Dataset

• Dataset: All-Age-Faces Dataset

---

⚙️ Model Details

• Architecture: Modified MobileNetV2 (Pretrained)
• Task: Age Regression
• Training Data Split: 80% train / 20% validation
• Epochs: 30
• Loss Function: Mean Squared Error (MSE Loss)
• Optimizer: AdamW with differential learning rates
  • Backbone: lr = 1e-4
  • Classifier: lr = 1e-3
• Learning Rate Scheduler: StepLR (decay every 10 epochs)
• Data Augmentation: Uses RandomResizedCrop, RandomHorizontalFlip, RandomRotation, and ColorJitter for better generalization.
• Fine-Tuning: Extended fine-tuning by freezing only the initial layers (i < 10) of MobileNetV2.

---

📊 Performance

Metric	Result
Mean Absolute Error (MAE)	5.76 years

(Evaluated on the validation set after 30 epochs.)

---

🚀 Usage

The trained model is available in the following formats for seamless integration into your mobile or web applications:

• PyTorch (age_sense.pt or one_a_complete_age_model.pth)
• ONNX (age_sense.onnx)
• CoreML (age_sense.mlmodel & age_sense_quantized.mlmodel)

---

📜 License

• Author: ray@jejememe
• License: MIT License

---

📮 Contact

For inquiries, please contact:

• Email: ray@jejememe.com
• GitHub: jejememe