A pre-configured instant-ngp workspace that includes helpful scripts for getting started with NVIDIA's NeRF model.
This repo will not hold your hand through the process, it just makes the process much easier.
-
- Ensure you have Python and ffmpeg installed (ffmpeg as an environment variable, see this tutorial on how to create a path environment)
- Install opencv-python:
pip install opencv-python
- Install instant-ngp and colmap either manually or by using the pre-made installation scripts found in
./install_scripts
- Add the downloaded COLMAP folder to your path environment variables, for example:
C:\COLMAP-3.7-windows-cuda
-
- Put your video inside the
./workspace/dataset_creation
folder and rename it toinput.mp4
- Run each script in the order by number (feel free to modify any if needed) or run
run_all.bat
to automate the process (this will make it harder to debug) - If successful, you will be left with a new folder named
dataset
. It should contain two things,transforms.json
and_images
. If both are present then congrats! You successfully created a dataset forinstant-ngp
- Run
instant-ngp.exe
(which you installed earlier) and drag 'n drop your dataset - I have also made some datasets in which you can try, they are included inside of the
dataset_examples
folder
- Put your video inside the