Adrian Albert / Nan Zhou Contact: aalbert@lbl.gov
Information on the built environment is crucial for a wide variety of urban science applications, including energy efficiency benchmarking, environmental resilience assessments and risk mitigation, and urban planning. Much of this information – e.g., building height, geometry, age, height of the first floor etc. – is only very sparsely available in most cities, primarily because of the significant cost and effort in collecting and maintain ing such databases.
This project aims to develop an AI-based approach for cross-view inference of building attributes from two complementary views of urban infrastructure: overhead (satellite imagery) and ground-level (such as from Google Street View). The satellite data is generally available world-wide, albeit at different spatial resolutions, whereas ground-level imagery is typically only available in developed countries. This observational data can be combined with labeled data from GIS catalogs on building attributes (such as tax assessors databases), where available. To complement the unavailability of ground-level imagery with associated labels about the attributes of interest, we will explore the use of realistic game engines (such as Unity or GTA). We will explore the use of Generative Adversarial Networks for learning the complex relationships between this cross-view data and “geographic functions” such as building height or age) that can be inferred from the data.
Desired skills:
– Comfortable with Python programming (object-oriented programming, numpy, matplotlib, pandas etc.)
– Familiarity with modern machine learning frameworks such as PyTorch or TensorFlow
– Knowledge and familiarity with modern machine learning, including deep learning. At a minimum, class work and projects, but previous research work using/developing deep learning methods preferred (ideally having worked with GANs)
-Interest in scientific applications of AI including one or more of climate, energy, materials