Level 1: Project Manipulation#

Projects are designed to facilitate the use of Datumaro CLI commands for complex tasks. If you need more information about the project’s concept and meaning, please refer to Projects. If you plan to perform various dataset management tasks using the Datumaro Python API instead of CLI commands, you may be able to skip this page. However, even when using the Python API, you will need to refer back to the project if you require dataset versioning.

Create a project#

The following command creates a project in the current working directory.

cd <path/where/you/want>
datum project create

Import a dataset to the project#

After creating the project, you can import datasets to the project. In this example, we import Cityscapes dataset to the project with naming it my-dataset.

datum project import -n my-dataset --format cityscapes -p <path/to/project> <path/to/cityscapes>

Remove a dataset from the project#

Conversely, it is also possible to delete a dataset that has been added to the project. This command removes the Cityscapes dataset named my-dataset from the previous step.

datum project remove my-dataset

Add model to project#

You can add an AI model into a project. The model requires an inference launcher for its model format. Currently, we only support OpenVINO™ launcher. Here is an example to add an OpenVINO™ model to the project.

datum model add -n my-model -l openvino -- -d <path/to/model.xml> -w <path/to/model.bin> -i <path/to/interpreter.py>


In addition to entering the path to the model weights (-w WEIGHTS) and metafiles (-d DESCRIPTION), you must enter the interpreter file path (-i INTERPRETER) written in Python to interpret that model output as well.

An example of the interpreter (``ssd_mobilenet_coco_detection_interp.py``)

Remove model from project#

We can remove my-model model from the project as follows.

datum model remove my-model