This contains the datasets and codes for training the 3D line registration method described in : PlueckerNet: Learn to Register 3D Line Reconstructions, CVPR2021.
Please download Structured3D and Semantic3D and Apollo datasets.
Please put the downloaded files under the folder ./dataset
Pytorch=1.1.0 : (This is the version on my PC, but I think it also works on yours)
numpy
opencv
tensorboardX
easydict
logging
json
If you find missing Prerequisites, please Google and install them using conda or pip
Our model is implemented in Pytorch. All our models are trained from scratch, please run the training codes to obtain models.
For pre-trained models, please download. Under the folder of each dataset, there is a folder named preTrained and you can find it there.
Please put the downloaded pre-trained models under the folder ./output
Run:
python main_train.py
Run:
python main_test.py
If you have questions, please first refer to comments in scripts.
If you like, you can cite the following publication:
Liu, Liu, Hongdong Li, Haodong Yao, and Ruyi Zha. "PlueckerNet: Learn to Register 3D Line Reconstructions." arXiv preprint arXiv:2012.01096 (2020).
@article{liu2020plueckernet, title={PlueckerNet: Learn to Register 3D Line Reconstructions}, author={Liu, Liu and Li, Hongdong and Yao, Haodong and Zha, Ruyi}, journal={arXiv preprint arXiv:2012.01096}, year={2020} }
If you have any questions (NOT those you can find answers via Google), drop me an email (Liu.Liu@anu.edu.au)