
Thesis: Code Generation
Compose vision models using visual programming such that the models and generator can be optimized independently to solve compositional tasks.
I'm Josh, a recent Computer Science M.S. graduate from the University of Illinois Urbana-Champaign with a bachelor's from the University of California Berkeley. My Master's thesis and expertise spans computer vision, natural language processing, and reinforcement learning.
My research focuses on leveraging both vision and language to extract information from images. My thesis research composes vision models using generated code to solve complex tasks, like visual-question answering.
Compose vision models using visual programming such that the models and generator can be optimized independently to solve compositional tasks.
Program autonomous vehicle to track a human with a camera and provide control inputs to follow the human.
Augment road image segmentation to include lane lines using analytical methods.
Style transfer with a pyramid structure to produce high-res images.