Many natural vision systems respond to color bands beyond RGB, revealing otherwise invisible material signatures. Check out our work on hyperspectral colorization, infrared material sensing, and human color perception.

Our computational models explain jumping spider behavior under defocus and spatial biases in larval zebrafish optic flow responses. They also enable a family of highly-efficient depth and velocity cameras [focal split, COD, metalens, focal track, focal flow; dataset], reveal mathematical connections between depth photography and phase microscopy [TIE], and describe natural scene statistics of underwater motion [zebrafish].

Best Student Paper ECCV 2016, Best Demo ICCP 2018, patent licensed to Metalenz
[jumping spider media coverage 1 2 3 4 5 6 7, ~4M youtube views 8]
[zebrafish media coverage 1 2 3 4 5 ... total: 75 outlets, ~1.5M readers]

Physics-informed machine learning allows us to combine models of the physical world with data-driven priors to solve difficult inverse problems like galaxy shear estimation for cosmology and material classification from thermal video .  A combination of machine learning and classic image processing techniques reveal more information from medical imaging modalities including X-ray, photoacoustic, and optical coherence tomography.

We work with astronomers through the SkAI AI Institute for the Sky.

[galaxy media coverage 1 2 3 4]