Back to work
WDSS: Wavelet-Driven Super-Sampling
P1Featured
Collaborative research: real-time neural super-resolution for game rendering, with an Unreal Engine capture plugin (arXiv preprint).
Overview
Four-person final-year research project; co-authored paper published in Elsevier Computers & Graphics (also on arXiv). Built a wavelet-space neural super-resolution pipeline in PyTorch that upscales low-resolution rendered frames using HR G-buffers, temporal history, and BRDF-based decomposition. My contributions: the Fourier-mapped INR module for high-frequency detail reconstruction, feature-extraction and dataset pipeline, and inference-time benchmarking. Frames were captured through a custom Unreal Engine plugin (C++/C#).