Doctoral Thesis Oral Defense - Arjun Teh

— 1:00pm

Location:
In Person and Virtual - ET - Reddy Conference Room, Gates Hillman 4405 and Zoom

Speaker:
ARJUN TEH , Ph.D. Candidate, Computer Science Department, Carnegie Mellon University
https://www.cs.cmu.edu/~ateh/

Computational Lens Design

Contemporary lens design, therefore, presents a multifaceted optimization challenge. A typical compound lens system is characterized by both continuous parameters, such as surface shape and thicknesses, and discrete choices, such as the number of elements and types of material. This mixed discrete-continuous parameter space creates a complex design landscape where the performance of the lens is highly sensitive to all of the choices of parameters. This design space has been historically hard for designers to navigate, necessitating assistance from theoretical and computational tools that help guide the search for performant designs. Yet, even with these tools, design remains time consuming and requires a great deal of designer input. In parallel, the graphics community has developed methods for differentiable rendering, which enable gradient-based optimization of image based tasks. These methods have been successfully applied to a variety of problems and are a great candidate for application to lens design. However, there are key differences in lens design from general rendering that make directly applying these methods to lens design challenging.

The purpose of this thesis is to develop methods that leverage ideas and techniques from differentiable rendering to address the challenges of lens design, by developing a set of theoretical and computational tools tailored to the unique requirements of lens design. Firstly, we build a method for calculating the unbiased gradient of light throughput with respect to lens parameters, enabling the optimization of lens speed. Secondly, we devise Markov chain Monte Carlo (MCMC) method that combines gradient-based optimization of continuous parameters with discrete mutations that change the number of elements in a lens system. Lastly, we derive a constant memory method for calculating gradients of ray paths through gradient-index (GRIN) materials allowing for optimization of GRIN lenses

Thesis Committee
Ioannis Gkioulekas (Co-chair)
Matthew O'Toole (Co-chair)
James McCann
Bernd Bickel (ETH Zürich)

In Person and Zoom Participation.  See announcement. 

For More Information:
matthewstewart@cmu.edu

Add event to Google
Add event to iCal