Normal Field Learning

Torch installation

python 3.8 recommended, Choose one of the following cuda versions

CUDA 11.3

Pytorch installation

pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu113

Pytorch scatter installation

pip install torch-scatter -f https://data.pyg.org/whl/torch-1.12.1+cu113.html

CUDA 11.7

Pytorch installation

pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117

Pytorch scatter installation

pip install torch-scatter -f https://data.pyg.org/whl/torch-1.13.1+cu117.html

CUDA 10.2

Pytorch installation

pip install torch==1.12.1+cu102 torchvision==0.13.1+cu102 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu102

Pytorch scatter installation

install torch-scatter -f https://data.pyg.org/whl/torch-1.12.1+cu102.html

See https://data.pyg.org/whl/ for supported version of pytorch-CUDA

Common installation

DVGO installation

git clone git@github.com:twjhlee/Normal-Field-Learning.git
pip install -r requirements.txt

OpenEXR installation

pip install git+https://github.com/jamesbowman/openexrpython.git

Surface normal estimator

All implementations are based on Bae, G., Budvytis, I., & Cipolla, R. (2021). Estimating and exploiting the aleatoric uncertainty in surface normal estimation. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 13137-13146).

Download https://drive.google.com/file/d/1pNZ1-4iX3o4bzCkd3k5GzkVl6ae9M7eo/view?usp=drive_link to surface_normal_uncertainty/experiments/nomask_noset1/models

Mask estimator

All implementations are based on Chen, L. C., Papandreou, G., Schroff, F., & Adam, H. (2017). Rethinking atrous convolution for semantic image segmentation. arXiv preprint arXiv:1706.05587.

Download https://drive.google.com/file/d/1fH2TH6jKCkmgcNVKFj0LtpX7RbU46P_U/view?usp=drive_link to deeplab/checkpoints

Example dataset

Download https://drive.google.com/file/d/1FN5Nbfv1RQ5-H-qW_bmVI-t2w-_DKHts/view?usp=drive_link to data

Usage

Preprocessing dataset

python preprocess.py --input_dir "path_to_input_dir"

Training, testing

python run_all.py --config configs/inmc/nfl_example/norm_kappa_ber_fixate.py

Pointcloud(ply) conversion

python tools/vis_volume.py "path to scene_mesh.npz" "threshold[0, 1]"