SPLATNet: Sparse Lattice Networks for Point Cloud Processing (CVPR2018)
Copyright (C) 2018 NVIDIA Corporation. All rights reserved. Licensed under the CC BY-NC-SA 4.0 license (https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode).
@inproceedings{su18splatnet,
author={Su, Hang and Jampani, Varun and Sun, Deqing and Maji, Subhransu and Kalogerakis, Evangelos and Yang, Ming-Hsuan and Kautz, Jan},
title = {{SPLATN}et: Sparse Lattice Networks for Point Cloud Processing},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages = {2530--2539},
year = {2018}
}
Install Caffe and bilateralNN
Note that our code uses Python3.
docker pull suhangpro/caffe:bpcn
You can also build this image with the Dockerfile.Include the project to your python path so imports can be found, e.g.
export PYTHONPATH=<PATH_TO_PROJECT_ROOT>:$PYTHONPATH
Download and prepare data files under folder data/
See instructions in data/README.md.
Usage examples
cd exp/facade3d
./dl_model_facade3d.sh # download pre-trained model
SKIP_TRAIN=1 ./train_test.sh
Prediction is output at pred_test.ply
, with evaluation results in test.log
.cd exp/facade3d
./train_test.sh
cd exp/shapenet3d
./dl_model_shapenet3d.sh # download pre-trained model
./test_only.sh
Predictions are under pred/
, with evaluation results in test.log
.cd exp/shapenet3d
./train_test.sh
We make extensive use of bilateralNN, which is proposed in these publications: