Deep Learning Architectures for Building Architecture

Title: Building Deep Learning Architectures to Understand Building Architecture Styles

Authors: Caroline Ho & Cole Thomson {cho19, colet}@stanford.edu

Course: CS 230 – Deep Learning

Requirements

• PyTorch and libraries: Anaconda Python distribution
• Dataset: Xu et al. 25-class architecture dataset

classify.ipynb

• Install tabulate: pip install tabulate
• Install TNT: pip install torchnet

Description

classify.ipynb

This notebook uses transfer learning to classify images of buildings by architectural style.

Best Results: After pretraining a DenseNet on ImageNet, we achieve an accuracy of 0.795833 and a F1 score of 0.789431. (Visualizations available in notebook.)

generate.ipynb

This notebook generates images of buildings conditioned on architecture styles using a conditional GAN.

Results after 20 epochs:

Our most successful generated image is this example of Ancient Egyptian architecture, which is visibly a pyramid:

However, most of our images, including this example of American Craftsman architecture, are less clear. (If you look closely, you can see a blurry gabled brown roof and white walls.)

Acknowledgments

Much of our code has been adapted from the following sources.

• Data: Architectural Style Classification using MLLR
• Classification: PyTorch Transfer Learning Tutorial and Finetuning Torchvision Models Tutorial
• Confusion Matrix: scikit-learn
• Conditional GAN: togheppi's cDCGAN