Skip to content

GaTech-RL2/nvf_cvpr24

Repository files navigation

Neural Visibility Field for Uncertainty-Driven Active Mapping

CVPR 2024

Shangjie Xue, Jesse Dill, Pranay Mathur, Frank Dellaert, Panagiotis Tsiotras, Danfei Xu

Installation

NVF has been tested on Ubuntu 20.04 with an RTX 3090, CUDA >= 11.7, python 3.10, and specific versions of nerfacc & viser. Below is line-by-line how we setup our environment dependencies.

conda create -y -n nvf python=3.10
conda activate nvf

conda env config vars set CUDA_HOME=$CONDA_PREFIX
conda env config vars set CUDA_INCLUDE_DIRS=$CONDA_PREFIX
conda deactivate; conda activate nvf

python -m pip install --upgrade pip
pip install torch==2.0.1+cu117 torchvision==0.15.2+cu117 --extra-index-url https://download.pytorch.org/whl/cu117

pip install ninja
conda install -y -c "nvidia/label/cuda-11.7.1" cuda-nvcc
conda install -y -c "nvidia/label/cuda-11.7.1" cuda-toolkit
pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

pip install --upgrade pip setuptools
pip install -e .

pip install  git+https://github.com/nerfstudio-project/viser.git@aa417815bf248ba15ee6e22cd4bb49bbc149dee8

pip uninstall -y nerfacc
pip install git+https://github.com/KAIR-BAIR/nerfacc.git@433130618da036d64581e07dc1bf5520bd213129

pip install -r nvf/requirments.txt 
pip install pytorch3d -f  https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/py310_cu117_pyt201/download.html

# for image generation
sudo snap install blender --classic

Usage

To evaluate NVF on the hubble scene, run

python eval.py --scene hubble --method nvf

You can run with different scenes:

python eval.py --scene {hubble,room,lego,hotdog} --method nvf

To train on the datasets, download the .blend and .ply files from our google drive and place them into /data/assets/blend_files/

Viewing

To view the NVF results during the training process, we use the viewer provided by nerfstudio. After eval.py is executing, an http url should be provided in the terminal. To view the viewer over SSH, see nerfstudio's guide. By default, the viewer shows the RGB visualization of the scene. To view the uncertainty, select the output render box, it currently should say rgb. Choose entropy from the dropdown options.

Configuration

To run NVF and baseline methods in different configurations, see all possible options with

python eval.py -h

Codebase

This repo builds upon nerfstudio and nerf_bridge. We use nerfstudio's implementation of Instant-NGP. Our modifications to Instant-NGP can be found in models/instant_ngp.py.

The ground truth training image is rendered through Blender. For adding new scenes for evaluation, take at look at config.py and nvf/env/Scene.py.

Citation

If you find this repo useful for your research, please consider citing our paper

@inproceedings{xue2024neural,
  title={Neural Visibility Field for Uncertainty-Driven Active Mapping},
  author={Xue, Shangjie and Dill, Jesse and Mathur, Pranay and Dellaert, Frank and Tsiotras, Panagiotis and Xu, Danfei},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={18122--18132},
  year={2024}
}

About

CVPR24: Neural Visibility Field for Uncertainty-Driven Active Mapping

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •