Architecture ex Machina

Show and Tell: What are you working on right now with AI/ML?

This is a thread to share finished work, work in progress, and anything else that you’ve created with a focus on AI and ML for the architecture / engineering / construction industry. We’d love to to see all the unique ways that firms across the spectrum are taking advantage of this new technology to perform interesting and valuable work.

No need to be bashful, even if you’re just starting down the path of AI/ML let us know what you’re trying to do and we can all chime in to provide guidance and feedback.

We have been focusing on the problem of acquiring the data necessary to train deep neural networks in the context of AEC work. At ACADIA 2018, we published a paper of some recent results showing that it is possible to train neural networks to recognize engineering symbols in real world drawings simply by generating data.

Performing this kind of semantic segmentation for relevant symbols is the foundation of any kind of higher understanding in machine learning. Deepmind’s recent developments also depend on a variation of this technique using disentangled representations to learn how to understand the spatial composition of toy-like 3D scenes:

did this 2 years ago:
https://flatspace.xyz
afaik this was the first automatic bitmap to 3d floor plan converter demonstrated in public
and here is a cute video showing how semantic inference improves with the training:
https://www.youtube.com/watch?v=GPXzyUt2iYA

Very impressive, thanks for sharing!

Is this focused on residential projects or can it create visualizations for arbitrary industry verticals?