Deep learning and classical image processing.
Built and trained custom UNet across 20 epochs to generate MNIST digits via DDPM. Leveraged DeepFloyd for text-conditioned generative image manipulation including inpainting, visual anagrams, and hybrid image synthesis.
View Project
Tested three CNN architectures for facial keypoints prediction: un-noised training set, pixel-noised with random transformations, and noised training set of 6000+ images.
View Project
Transformed 2D shapes into 3D puffball models using Medial Axis Transform. Leveraged Delaunay triangulation to convert it to a mesh for surface reconstruction.
View Project
Implemented homography transformation for novel view synthesis and image mosaic'ing, maintaining 3-pixel Euclidean distance between 150 corresponding keypoints.
View Project
Used Delaunay triangulation, geometric morphing, and color interpolation for seamless face transformations.
View Project
Explored filtering techniques to produce Gaussian and Laplacian stacks for image blending.
View Project
Transformed black and white photographs into vibrant color images with a Normalized Cross Correlation alignment algorithm.
View Project