Architectural Style Classification
Python TensorFlow Keras NumPy scikit-learn
Description
An image classification project for international architectural styles using deep learning. Images are classified into 7 classes from the International Architectural Styles dataset (Kaggle), with a total of 1,938 images.
| Class | Images |
|---|---|
| Maya architecture | 337 |
| Ottoman architecture | 324 |
| Minangkabau architecture | 315 |
| Architecture of the Qing Dynasty | 293 |
| Brutalist architecture | 256 |
| Dravidian architecture | 241 |
| Balinese architecture | 172 |
The model uses MobileNetV2 (pretrained on ImageNet) as the base with additional Conv2D and MaxPooling2D layers, trained in two phases:
- Feature extraction: base model frozen, only top layers trained
- Fine-tuning: last 30 layers of MobileNetV2 unfrozen with a lower learning rate
Test Accuracy: 86.25%