A single index for the semester. Each page includes an overview, implementation details, and visual results. Projects progress from camera geometry → classical vision → neural representations → generative models.
Three mini-studies on perspective, focal length, and the center of projection: stepped-back zoom portraits, street perspective compression, and a dolly zoom (Vertigo shot) built from stills.
Align B/G/R plates into modern RGB photos using binarized high-pass features scored with NCC, refined with a coarse-to-fine pyramid for full-resolution plates.
Convolution from scratch, finite-difference gradients, DoG, unsharp masking, hybrid images, and multi-resolution blending (Gaussian/Laplacian stacks + mask pyramids).
A combined write-up for 3A + 3B: manual correspondences → homography → inverse warping, and an end-to-end automatic pipeline with corners, descriptors, ratio-test matching, RANSAC homography, warp + blend.
From raw phone images to a learned continuous 3D representation: calibration/poses, a 2D neural field warm-up, training NeRF on Lego, and a full NeRF of my own object via volumetric rendering.
Two views of the same goal: noise → image. In 5A, implement sampling + CFG + editing (SDEdit/inpainting). In 5B, train UNets (single-step denoising + flow matching velocity fields) with time/class conditioning and CFG.