Analyze the Selection Results

Lightly Platform

You can view and explore the dataset interactively on the Lightly Platform.


Local Storage

If you are using local storage and want to visualize the selected data in the Lightly Platform, you need to use thelightly-serve command. Point the input_mount and the lightly_mount to the same directories you mounted to the Docker container. More information on lightly-serve can be found in the chapter on Local Storage.

lightly-serve input_mount=/MY/PATH/TO/INPUT/DIRECTORY lightly_mount=/MY/PATH/TO/WORKING/DIRECTORY

Embedding view of a dataset in the Lightly Platform.

Report and Other Artifacts

Each Lightly Worker run creates artifacts containing useful information about the run which can be used to get further insights. One of the artifacts generated with every run is the report PDF file. You can download it like this:

from lightly.api import ApiWorkflowClient

# Create the Lightly client to connect to the API.
client = ApiWorkflowClient(token="MY_LIGHTLY_TOKEN", dataset_id="MY_DATASET_ID")

# Get all runs for a given dataset sorted from old to new.
runs = client.get_compute_worker_runs(dataset_id=client.dataset_id)
run = runs[-1] # get the latest run

# Download the report.pdf
client.download_compute_worker_run_report_pdf(run=run, output_path="my_run/artifacts/report.pdf")

Your first report will contain:

  • Information about the selected and discarded images and videos.
  • 2D scatter plots of the embeddings of the images with thumbnail overlays.
  • Pairwise distance metrics comparing image similarity before and after running the Lightly Worker.
  • Metadata information about the images and videos in the dataset.

If you choose more complicated selection strategies involving predictions or active learning the report will cover these aspects as well.

Exempt from a report of a Lightly Worker run on a subset of the Berkley DeepDrive dataset.

Exempt from a report of a Lightly Worker run on a subset of the Berkley DeepDrive dataset.

What’s Next