CS 8803 A3B – NeRF Lego Render – Ethan Villalovoz

Assignment 3B – Neural Radiance Fields (NeRF)
CS 8803/4803: Computer Graphics in the AI Era
Georgia Tech

This video demonstrates my implementation of a simplified Neural Radiance Field (NeRF) pipeline using the tiny_lego dataset.

Step 1 – NeRF Model Implementation:
• Positional encoding of 3D coordinates using sinusoidal frequency functions
• Volume rendering with transmittance-based front-to-back compositing
• Network training using MSE loss
• Evaluation using PSNR

The PyTorch implementation samples points along camera rays, predicts RGB and density values, applies Sigmoid activation to color and ReLU to density, and integrates radiance using the volumetric rendering equation.

Step 2 – GLSL Rendering:
The trained NeRF model was serialized and implemented in fragment.glsl. The shader performs ray marching, queries the neural network per sample point, and accumulates color using transmittance-based compositing to synthesize novel views in real time.

Author: Ethan Villalovoz

Видео CS 8803 A3B – NeRF Lego Render – Ethan Villalovoz канала Ethan Villalovoz