This repository has been built over this pytorch implementation of NeRF: https://github.com/yenchenlin/nerf-pytorch
- pytorch
- torchvision
- numpy
- imageio
- imageio-ffmpeg
- matplotlib
- configargparse
- opencv-python
- pyquaternion
Run the following commands.
cd inerfpip install -r requirements.txt
Implementation Details:
Currently, this implementation supports 2 datasets i.e. nerf-synthetic and nerf-llff. Performing pose estimation
for any query image of a scene also requires its pretrained NeRF representation. This repository may not contain
all the details to train NeRF. We have evaluated results for scenes lego (synthetic) and fern (llff).
run_inerf.py script performs pose estimation for a random query image from the validation/test dataset.
The pose is randomly initialized in some vicinity of the ground truth camera pose.
Please check out the report for more details.
It usually takes less than 5 minutes to optimize pose for 1 query image on a single GPU.
Data:
- lego (synthetic): pretrained NeRF, images and corresponding camera poses
- fern (llff): pretrained NeRF, images and corresponding camera poses
How to optimize camera pose:
python run_inerf.py --config inerf_configs/{SCENE}.txt (replace {SCENE} with lego | fern)
Also, for more data, pretrained models and other implementation details on NeRF, checkout the file nerf_README.md
Code Citations:
- Pytorch NeRF implementation: https://github.com/yenchenlin/nerf-pytorch
- Modern Robotics for exponential parameterization: https://github.com/NxRLab/ModernRobotics/blob/master/packages/Python/modern_robotics/core.py