Project 6
NeRFs
Implemented Neural Radiance Fields (NeRF) to reconstruct 2D and 3D representations of images and spaces. For 2D, optimized an MLP with sinusoidal positional encoding to fit image data. For 3D, extended to multi-view synthesis using ray sampling, volume rendering, and a modified MLP to predict densities and colors for 3D points.
Project 5
Diffusion Models
Explored the power of diffusion models by implementing denoising sampling loops, image inpainting, and text-conditional image transformations. Trained custom diffusion models from scratch on MNIST to iteratively generate and enhance images with time and class conditioning, achieving high-quality outputs with UNet-based architectures and classifier-free guidance techniques.
Project 4
Image Warping + Mosaicing / Feature Matching for Autostitching
Developed an image mosaicing pipeline involving image alignment, homography computation, and warping to seamlessly blend multiple images into cohesive mosaics. Enhanced with automated feature matching using Harris corners, feature descriptors, and RANSAC for robust stitching. Results include both manually and automatically stitched panoramas.
Project 3
Face Morphing
Created a face morphing pipeline to generate smooth transitions between faces, compute the average face of a population, and produce caricatures by extrapolating from the population mean. This project combines shape warping and color blending to explore facial transformations and visual effects.
Project 2
Filters and Frequencies
Explored image processing techniques such as edge detection, image sharpening, hybrid image creation, and multiresolution blending. Applied advanced methods like Gaussian and Laplacian stacks to seamlessly blend images.
Project 1
Colorizing the Prokudin-Gorskii photos
Aligned and colorized digitized Prokudin-Gorskii glass plate images using image processing techniques to produce vibrant RGB images with minimal visual artifacts.