Projects
3D Image Registration
Advised by Prof. Laura Waller and graduate students Leyla Kabuli, Ruiming Cao, and Tiffany Chien, implemented and validated a Multi-Layer Perceptron (MLP) based model for 3D medical image registration, with a focus on defining the neural network as a space-time MLP and a motion MLP, followed by memory optimization.
Light Field 3D Reconstruction
Worked on light field 3D reconstruction algorithm based on mobile captured images for Kuaishou Flagship, a short video social networking platform with 100M+ users, boosting modeling efficiency from 1 week to 4-6 hours. Created face tracking 3D filters and developed the UI to streamline object capturing and model generation workflow.
Vision Correcting Display
Advised by Prof. Brian Barsky, designed, prototyped, and characterized light field display algorithms to enable convolutional prefiltering of vision correcting displays that pre-distorts the presented content for visually impaired people through the integration of hardware and software.
Face Detection
Used neural networks to automatically detect facial keypoints. I chose ResNet18 for facial keypoints detection and UNet for Pixelwise Classification.
Face Morphing
Implemented face morphing algorithm where morphing = warping + cross-dissolving. Warping was done by performing Delaunay Triangulation and applying affine transformations to input images, with different parts (triangles) of being warped and inversely warped accordingly so that we can change the geometries. Calculated the average face of a population, and created a growth video of myself!