Project Abstract

This project presents a novel deep learning method for generating high-quality Global Illumination (GI) from a simple Direct Illumination (DI) input. Our approach utilizes a lightweight and efficient neural network, which directly synthesizes complex lighting effects like color bleeding and soft shadows in a single forward pass. A key achievement of this work is its practicality: the model is specifically designed with a compact architecture, enabling it to run smoothly on consumer-grade GPUs with as little as 12GB of VRAM. This makes photorealistic rendering more accessible without the need for expensive hardware or time-consuming path tracing.