Sibeak Lee, Kyeongsu Gang, Hyeonwoo Yu
We present the Bayesian Neural Radiance Field (NeRF), which explicitly quantifies uncertainty in geometric volume structures without the need for additional networks, making it adept for challenging observations and uncontrolled images.
The results have been adjusted to account for the uncertainty in (Far Left) Base, (Left-Center) Color, (Center) Density, (Right-Center) Density and Color, and (Far Right) Occupancy.
- Synthetic data (Blender) and real-world data (LLFF) : NeRF dataset.
- Modelnet Dataset : ModelNet dataset
Our proposed method excels in scenarios with limited data, hence we recommend downloading the dataset and appropriately processing it for experimentation.
conda create --name bayesian_nerf python=3.8
conda activate bayesian_nerf
pip install -r requirements.txt
cd NeRF_for_rgb_img
cd NeRF_baseline
python run_nerf.py --config configs/synthetic.txt --expname ../../result/chair/4_baseline --datadir ./data/nerf_synthetic/chair_4
cd ..
cd NeRF_color
python run_nerf.py --config configs/synthetic.txt --expname ../../result/chair/4_color --datadir ./data/nerf_synthetic/chair_4
cd ..
cd NeRF_density
python run_nerf.py --config configs/synthetic.txt --expname ../../result/chair/4_density --datadir ./data/nerf_synthetic/chair_4
cd ..
cd NeRF_density_and_color
python run_nerf.py --config configs/synthetic.txt --expname ../../result/chair/4_den_col --datadir ./data/
cd ..
cd NeRF_occupancy
python run_nerf.py --config configs/synthetic.txt --expname ../../result/chair/4_occupancy --datadir ./data/
cd ..
cd ..
- RGB img
cd NeRF_for_rgb_img
cd [Method Name]
python run_nerf.py --config configs/synthetic.txt --expname <Output Path> --datadir <Dataset Path>
cd ..
cd ..
- Depth img
cd NeRF_for_depth_img
python <Method.py> --config configs/coarse.txt --expname <Output Path> --datadir <Dataset Path>
cd ..