thabsheerjm / nerf

Neural Radiance Fields - generating new perspectives of intricate scenes by refining a continuous volumetric scene representation through a limited selection of input views.

NeRF - Neural Radiance Field

Implementing the original method from [1]. Neural Radiance Fields (NeRF) is a novel approach for synthesizing 3D scenes from 2D images. It represents scenes as continuous 5D functions that output color and density for any given 3D point in space and 2D viewing direction. This method allows for high-quality, photorealistic rendering of 3D scenes from novel viewpoints, not just those captured by the input images. NeRF achieves this by training a neural network with multiple 2D images of a scene taken from various viewpoints, learning the scene's structure and appearance. The network then predicts the color and opacity for any given point and viewing angle, enabling the creation of new images from perspectives not originally photographed. For a detailed explanation, please see the project page.

How to Run

Train the network:

python train.py

Test the network:

python test.py

Results

Acknowledgement

This project was part of RBE549-Computer Vision (Fall 22) at Worcester Polytechic Institute[1].
Team members :Thabsheer Machingal and Krishna Madhurkar

References

[1] NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis
[2] RBE549-Project3