Code of DANCE: Domain Adaptation of Networks for Camera Pose Estimation: Learning Camera Pose Estimation Without PoseLabels by Jack Langerman, Ziming Qiu, Gabor Soros, David Sebok, Yao Wang, Howard Huang Nokia Bell Labs and New York University, 2020
Paper link: arxiv
Dataset link: Dataport. In the dataset, we have training images with 100,000 labeled rendered images and 28411 unlabeled real camera images. We also have the validation set (1637 labeled real camera images) and test set (2104 labeled real camera images).
Training:
(1) run histogram_match.ipynb
to preprocess the training rendered images.
(2) going into cut folder, use the prepare_dataset.ipynb
to prepare training data, then run run_py_job.sbatch
to train the CUT GAN model.
(3) run train_init_scr_cut.ipynb
to train the final scene coordinate regression model.
(1) run test.ipynb
Our work:
@article{DBLP:journals/corr/abs-2111-14741,
author = {Jack Langerman and Ziming Qiu and G{\'{a}}bor S{\"{o}}r{\"{o}}s and D{\'{a}}vid Sebok and Yao Wang and Howard Huang},
title = {Domain Adaptation of Networks for Camera Pose Estimation: Learning
Camera Pose Estimation Without Pose Labels},
journal = {CoRR},
volume = {abs/2111.14741},
year = {2021},
url = {https://arxiv.org/abs/2111.14741},
eprinttype = {arXiv},
eprint = {2111.14741},
timestamp = {Wed, 01 Dec 2021 15:16:43 +0100},
biburl = {https://dblp.org/rec/journals/corr/abs-2111-14741.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
The code is based on CUT and PoseNet:
@inproceedings{10.1007/978-3-030-58545-7_19,
author="Park, Taesung and Efros, Alexei A. and Zhang, Richard and Zhu, Jun-Yan",
title="Contrastive Learning for Unpaired Image-to-Image Translation",
booktitle="Computer Vision -- ECCV 2020",
editor="Vedaldi, Andrea and Bischof, Horst and Brox, Thomas and Frahm, Jan-Michael",
year="2020",
publisher="Springer International Publishing",
address="Cham",
pages="319--345",
isbn="978-3-030-58545-7"
}
@inproceedings{7410693,
author={Kendall, Alex and Grimes, Matthew and Cipolla, Roberto},
booktitle={2015 IEEE International Conference on Computer Vision (ICCV)},
title={PoseNet: A Convolutional Network for Real-Time 6-DOF Camera Relocalization},
year={2015},
pages={2938-2946},
doi={10.1109/ICCV.2015.336}
}