DGP Command-Line Interface

dgp/cli.py is the main CLI entrypoint for handling DGP datasets.

Visualize DGP-compliant Scene and SceneDataset

DGP CLI subcommands visualize-scenes and visualize-scene can be used to visualize DGP-compliant data.

  • To visualize a split of a **DGP SceneDataset**, run python dgp/cli.py visualize-scenes:

    Show the help message via:

    dgp$ python dgp/cli.py visualize-scenes --help

    Example command to visualize the images from CAMERA_01, CAMERA_05, CAMERA_06 and point cloud from LIDAR along with ground_truth annotations bounding_box_2d, bounding_box_3d from train split of the toy dataset tests/data/dgp/test_scene/scene_dataset_v1.0.json, and store the resulting videos in --dst-dir vis. One can find the resulting 3D visualization videos in vis/3d and 2D visualization videos in vis/2d.

    dgp$ python dgp/cli.py visualize-scenes --scene-dataset-json tests/data/dgp/test_scene/scene_dataset_v1.0.json --split train --dst-dir vis -l LIDAR -c CAMERA_01 -c CAMERA_05 -c CAMERA_06 -a bounding_box_2d -a bounding_box_3d

    3d-viz

Add flag render-pointcloud to render projected pointcloud onto images:

dgp$ python dgp/cli.py visualize-scenes --scene-dataset-json tests/data/dgp/test_scene/scene_dataset_v1.0.json --split train --dst-dir vis -l LIDAR -c CAMERA_01 -c CAMERA_05 -c CAMERA_06 -a bounding_box_2d -a bounding_box_3d --render-pointcloud

3d-viz-proj

  • To visualize a single **DGP Scene**, run python dgp/cli.py visualize-scene:

    Show the help message via:

    dgp$ python dgp/cli.py visualize-scene --help

    Example command to visualize the images from CAMERA_01, CAMERA_05, CAMERA_06 and point cloud from LIDAR along with ground_truth annotations bounding_box_2d, bounding_box_3d from the toy Scene tests/data/dgp/test_scene/scene_01/scene_a8dc5ed1da0923563f85ea129f0e0a83e7fe1867.json, and store the resulting videos in --dst-dir vis. One can find the resulting 3D visualization video in vis/3d and 2D visualization video in vis/2d.

    dgp$ python dgp/cli.py visualize-scene --scene-json tests/data/dgp/test_scene/scene_01/scene_a8dc5ed1da0923563f85ea129f0e0a83e7fe1867.json --dst-dir vis -l LIDAR -c CAMERA_01 -c CAMERA_05 -c CAMERA_06 -a bounding_box_2d -a bounding_box_3d

Coming soon: Retrieve information about an ML dataset in the DGP

DGP CLI provides information about a dataset, including the remote location (S3 url) of the dataset, its raw dataset url, the set of available annotation types contained in the dataset, etc. For more information, see relevant metadata stored with a dataset artifact in DatasetMetadata and DatasetArtifacts.

dgp$ python dgp/cli.py info --scene-dataset-json <scene-dataset-json>

Coming soon: Validate a dataset

DGP CLI provides a simplified mechanism for validating newly created datasets, ensuring that the dataset schema is maintained and valid. This is done via:

dgp$ python dgp/cli.py validate --scene-dataset-json <scene-dataset-json>