C-NERF: Representing Scene Changes as Directional Consistency Difference-based NeRF

Installation

git clone https://github.com/C-NeRF/C-NeRF.git
cd C-NeRF
conda create -n cnerf python=3.7
conda activate cnerf
pip install -r requirements.txt

Download NeRF models

You can download the pre-trained models from drive. Unzip the logs to the project root dir to save training time.

Download Demo Dataset

You can download the demo datasets from drive. Unzip the downloaded data to the project root dir in order to train.

Obtain CD Results

First download trained NeRF Models and dataset. Then,

bash C-NeRF.sh

This command will obtain the render images with change markers, the results are saved to ./logs/chess1/results_path_199999

Train NeRF models before and after scene change

First download the dataset. Then,

python run_nerf.py --config configs/chess1.txt
python run_nerf.py --config configs/chess2.txt