rgbd_dataset

Dataset utilities for sequences of posed RGBD frames. Some of the gradslam data utilities were used as a reference.

Example Data

You can download the Replica dataset to using the download script from Nice-SLAM.

Save it to $DATA_PATH.

Example Usage

To visualize an entire sequence as a point cloud, try

python3 pcd_scene.py dataset=Replica dataset.base_path=$DATA_PATH dataset.scene=room0 voxel_size=0.025 dataset.sequence_stride=20

Parameters

Dataset configs are defined with Hydra.

All dataset classes are derived from BaseRGBDDataset, which can be used to load RGB, depth and camera poses. The following arguments can be added to the above command and other scripts using this repo, like Concept-Nodes.

• dataset.dataset_name=replica: A string describing the dataset.
• dataset.base_path=$DATA_PATH: Dataset path.
• dataset.scene=room0: The scene name. The class will try to load the data from base_path/scene.
• dataset.width=1200: Original width of the RGB and depth images.
• dataset.height=680: Original width of the RGB and depth images.
• dataset.resized_width=-1: Resized width of the RGB and depth images. Set to -1 to keep the original width.
• dataset.resized_height=-1: Resized width of the RGB and depth images. Set to -1 to keep the original height.
• dataset.sequence_start=0: Where to start loading frames. This can be used to skip the beginning of a trajectory.
• dataset.sequence_end=-1: When to end loading frames. This can be used to skip the end of a trajectory. Set to -1 to load data until the last frame.
• dataset.sequence_stride=1: Skip frames.
• dataset.relative_pose=true: Load relative camera poses. The first pose becomes the origin.
• dataset.depth_scale=6553.5: Depth scale.
• dataset.point_cloud=true: Unproject the RGBD to an Open3D point cloud.
• dataset.depth_trunc=8.0: Truncate the point cloud based on distance from camera. This only applies to the point cloud generated by the previous argument.

Look at conf/dataset/Replica.yaml to see how Replica is configured.

Additionally, the intrinsics can be configured with the fx, fy, cx and cy parameters or loaded from files, as done in the Scannetpp class.

Scripts

Look at the first level of conf for the configurations of the other scripts.