Takes in video, calculates a NeRF using nerfstudio, and generates novel view videos.
- setup.sh : shell script to test setup and to pull Docker image
- Problem_1.py : Python script to train NeRF from video
- Problem_2.py : Python script to use a NeRF to generate novel view video
- Problem_3.py : Python script to re-train a NeRF adding text instruction editing
- Run_All_Notebook.ipynb : Jupyter notebook to optionally run problems
- launch_docker.sh : convenient shell script to launch Docker with custom path mount
(Ubuntu 20.04 LTS)
https://docs.nvidia.com/datacenter/tesla/tesla-installation-notes/index.html#ubuntu-lts
The CUDA runtime is baked into the Docker image and should not be required on the host machine.
https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html
Docker Desktop allows for multiple contex and launching Docker through a remote machine. These multiple context will cause errors when requiring local admin privileges (through sudo
) to enable local GPU access.
https://docs.docker.com/desktop/install/ubuntu/
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html
This repo requires use of NeRF Studio through Docker image. It will not work with NeRF Studio pulled from another source.
https://docs.nerf.studio/en/latest/quickstart/installation.html#use-docker-image
docker pull dromni/nerfstudio:0.2.1
- Ubuntu OS:
- Linux 20.04 LTS
- Kernal: 5.15.0-69-generic
- Nvidia Hardware + Drivers
- GeForce RTX 2060
- CUDA Compute Capability: 7.5
- CUDA arch: 75
- NVIDIA Drivers: 530.30.02
- GCC: 9.4.0
- CUDA: 11.8.0-520.61.05-1
- Docker
- Docker: 23.0.6, build ef23cbc
- Docker Engine: 23.0.6
- Docker Compose: v2.17.3