On March 25th (5:00pm-9.30pm PST) , SFCDI welcomes Petr Mitev , San Francisco, CA 94107 Microsoft Reactor, 680 Folsom St #145.
Speaker: Petr Mitev Computational Design Leader at NBBJ
I’m an AEC polymath who thrives on diverse problems and solving them through an agile and experimental application of technology. My education and early career were entirely in step with “traditional” architecture pedagogy and values: earning BS and M-Arch degrees at the University of Cincinnati, while simultaneously working with firms in the US and abroad to build a portfolio of diverse experience. After graduating, I joined KieranTimberlake as an architect, and later moved to the Research Group to start the Design Computation Core as the firm’s first Computational BIM Leader. After several years there, I joined NBBJ in a similar role and in 2018 transitioned to lead the firm-wide NBBJ Design Computation Team. Currently, I’m focused on working with our Director of Digital Innovation to design & deploy custom software solutions which empower us and our partners to deliver the best possible designs to our clients and communities.
This workshop is a crash-course in accessible and practical machine learning, with a focus in computer-vision models. We’ll learn about popular ML models by building our own small IoT application which will use the models to generate data, and then we’ll dive into making our own simple predictive model from the gathered data to make predictions based on what our IoT sensor has collected.
Real-time Spatial Analysis – Using Google’s coco-ssd and mobile-net v2 models, we’ll extract information from images/videos that come from a simple IOT sensor camera (or a webcam for the purpose of the workshop) which can be used to inform design and/or a post-occupancy evaluation of a space. Some of the types of information we can try to extract are:
- How many people are in the space.
- Whether the space is highly trafficked, or it is a sedentary meeting spot.
- What other kinds of objects/entities are in the space.
- Whether any custom objects are present (type of furniture, plant, light, etc.)
We can then use the collected data to make predictions such as:
- When the space will be most/less occupied (when lighting and HVAC systems should be turned down/up).
- What kind of furniture distribution is optimal for the space (how much seating spaces, standing, meeting spots, informal gathering, etc.)?
Matthew Gilbert of ICAIR (remote) presenting “Layout Optimization of Building Structures” & new tool for Rhino/Grasshopper