Architectural Image Classifiers
Using design data to train a neural network
A convolutional neural network is trained on images of buildings from five common architectural styles, with a thousand images from each style. The images are distilled to their most relevant features, which are fed to individual “neurons” that use those features to predict the style of the image. Over a series of generations, a model emerges capable of predicting the architectural style of any given image.
A similar exercise with “starchitect” designs creates a model capable of finding similarities in new designs. This type of analysis, when applied across large datasets encompassing entire cities and regions, allows for broad understandings of where certain styles or features are represented and how that may impact the success of a proposed building.
A similar exercise with “starchitect” designs creates a model capable of finding similarities in new designs. This type of analysis, when applied across large datasets encompassing entire cities and regions, allows for broad understandings of where certain styles or features are represented and how that may impact the success of a proposed building.
Research (2018)
Next Project >