This repository presents a lightweight version of NeRF, which is trainable within 5 hours on Google Colab. Pre-trained checkpoints and Colab scripts are provided. It is useful for people new to NeRF to get a quick sense of what NeRF is.
The original model is hard to train on laptops or computers without strong GPUs. Mini-NeRF eliminate the inaccessibility of NeRF, anyone with a google account will have access to Colab and Mini-NeRF can be trained on free GPUs provided by Colab.
The training result can also be easily viewed on Colab, the visualization code cells are also included in the scripts.
blender_paper_lego_spiral_100000_rgb.mp4
An interactive visualization interface is also provided for user to view customized angles of the synthesized object.
- Clone this repo
git clone https://github.com/ChunHsuehLee/Mini-Nerf.git
- Upload the whole folder to your google drive.
- (Optional) Use the pre-trained checkpoints. Download the checkpoints here, upload the whole folder (include
logs
) to theMini-Nerf
folder you just uploaded. - Run the Colab script (.ipynb file) cell by cell.
- Have fun!
Thank yenchenlin and bmild for their super useful repos!