Parallel Python and GUI¶

You can run the web UI in the background and continue executing Python code from your main script. This is especially useful in notebooks where you want to keep exploring or updating the dataset after the GUI starts. You can explore the dataset interactively in the GUI, then decide to add some metadata or do selections in Python, and then continue exploring the updated dataset in the GUI without having to restart it.

The following example works best in a Jupyter notebook environment:

Cell 1: Setup¶

import lightly_studio as ls

Cell 2: Create a dataset from COCO data¶

dataset_path = ls.utils.download_example_dataset(download_dir="dataset_examples")
dataset = ls.ImageDataset.create("coco_instance_segmentation_dataset")
dataset.add_samples_from_coco(
    annotations_json=f"{dataset_path}/coco_subset_128_images/instances_train2017.json",
    images_path=f"{dataset_path}/coco_subset_128_images/images",
    annotation_type=ls.AnnotationType.INSTANCE_SEGMENTATION,
)

Cell 3: Start the GUI in the background¶

ls.start_gui_background()

Now you can click on the link printed in the output to open the GUI in your browser.

Cell 4: Run Python code while the GUI stays open¶

for i, sample in enumerate(dataset):
    sample.metadata["index_in_dataset"] = i

dataset.query().selection().metadata_weighting(
    n_samples_to_select=10,
    selection_result_tag_name="top_10_by_metadata_weighting",
    metadata_key="index_in_dataset",
)

After a refresh, you can see the updated metadata and selection in the GUI.

Cell 5: Stop the background GUI server¶

ls.stop_gui_background()

Notes:

Call start_gui_background() before interacting with the GUI and stop_gui_background() when you are done.