My dog, Chloe, in 3d.

ez-instant-ngp

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.

Tutorial

• Installation

  • 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

• Creating Datasets

  • Put your video inside the ./workspace/dataset_creation folder and rename it to input.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 for instant-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