Skip to content

Implementation of Sat-NeRF, a new NeRF architecture applied to satellite imagery for surface reconstruction

Notifications You must be signed in to change notification settings

fedesemeraro/surfnerf

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Surf-NeRF

This project implements the Shadow Neural Radiance (S-NeRF) field from this repository, starting from a modified PyTorch NeRF implementation. The model is able to generate novel views from a sparse collection of satellite images of a scene, as well as estimating a Digital Elevation Model (DEM) of the surface.

Dataset

The satellite image dataset that was used can be found at this link and it should be placed in a folder called "data" (e.g. data/068 for the images of JAX). The data can be augmented by modifying the inputs in the data_augmentation.py script and then running the command below. You can pass gauss=True to add gaussian blur in the augmented image along with sigma=0.2 to specify the radius of the gaussian blur.

python scripts/data_augmentation.py --gauss=True --sigma=0.2

Installation

It is recommended to create a conda environment using the following command from the root project folder:

conda env create
conda activate surfnerf

Then follow the instructions recommended on this website in order to install the correct version of PyTorch (CPU or GPU enabled).

How to run

To train NeRF on an example dataset run:

python run_nerf.py --config configs/068/068_config.txt

Project Contributors

  • Federico Semeraro fsemeraro6 AT gatech.edu
  • Yi Zhang yzhang3416 AT gatech.edu
  • Wenying Wu wwu393 AT gatech.edu
  • Patrick Carroll pcarroll7 AT gatech.edu

Cite

If you use Surf-NeRF in your research, please use the following BibTeX entries to cite our paper:

@misc{semeraro2023nerf,
      title={NeRF applied to satellite imagery for surface reconstruction}, 
      author={Federico Semeraro and Yi Zhang and Wenying Wu and Patrick Carroll},
      year={2023},
      eprint={2304.04133},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

About

Implementation of Sat-NeRF, a new NeRF architecture applied to satellite imagery for surface reconstruction

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages