Nerfstudio was a breeze to setup and works with CUDA version >= 11.8.

Opted for the Docker container which can be pulled down with the following command

docker pull dromni/nerfstudio:0.3.3

Once the ‘poster’ dataset is downloaded and training is started, it comes with a nice web visualizer that looks like the following:

After training completes (at 29999 steps), the trained model can be reloaded. A camera path can be defined for video rendering. The following videos shows the same camera path rendered using nerfacto (left) and instant-ngp (right) models.

References

Nerfstudio: A Modular Framework for Neural Radiance Field Development
M. Tancik, E. Weber, E. Ng, R. Li, B. Yi, J. Kerr, T. Wang, A. Kristoffersen, J. Austin, K. Salahi, A. Ahuja, D. McAllister and A. Kanazawa
ACM SIGGRAPH 2023 Conference Proceedings, 2023