Bell Labs robot garage for DANCE: Domain Adaptation of Networks for Camera Pose Estimation
This dataset is a supplement of the paper DANCE: Domain Adaptation of Networks for Camera Pose Estimation: Learning Camera Pose Estimation Without Pose Labels . The dataset contains a sample scene of a robot garage. The scene was captured by a Leica BLK360 laser scanner, and 16 scans were merged into a single point cloud of 118M colored points. The dataset also contains ~100k synthetically rendered images and scene coordinates generated form the point cloud. Furthermore, the dataset includes ~30k real photos of the space recorded by an iPhone 6 together with camera poses captured by a WorldViz system.
This dataset can be used together with the paper and code at https://github.com/JackLangerman/DANCE
Note (15.12.2021): The .zip upload of DataPort is currently broken, so we had to rename some file to .zip.dat. Please rename all files back to .zip after download.
The point cloud is in PTS ASCII format. The first line contains the number of points, other lines contain point data in X, Y, Z, I, R, G, B order.
Feel free to contact the authors for more information.
- train_labeled_rendered_poses.zip.dat (3.04 MB)
- test_set.zip.dat (848.06 MB)
- validation_set.zip.dat (648.75 MB)
- train_labeled_scene_coords.zip.dat (9.33 GB)
- train_unlabeled_real_images.zip.dat (8.52 GB)
- train_labeled_rendered_0.zip (10.95 GB)
- train_labeled_rendered_1.zip (10.94 GB)
- train_labeled_rendered_2.zip (10.93 GB)
- train_labeled_rendered_3.zip (10.93 GB)
- train_labeled_rendered_4.zip (10.93 GB)
- 200827_robot_lab_point_cloud.zip (1.42 GB)