How to create and work with slices in a model run.

A slice is a subset of data rows that share a common characteristic. You can construct a slice by using one or more filters to curate a collection of data rows. Often users will combine filters to surface high-impact data and then save the results as a slice.

Create a slice

Slices in model runs can be created either manually through filtering and hand-selecting data rows or, if model predictions are uploaded, can be auto-generated.



See this page to learn the limits for creating slices.

Manually create slices

Once you have narrowed down a subset of data rows using one or more filters or hand-selecting data rows, click Save slice to save the subset of data rows as a slice. You will be prompted to give a name (3 to 30 characters) and an optional description for your slice.

After you create a slice, the slice will show up in your list of Slices in the left side panel of the model run.

Auto-generated slices

After you upload model predictions to a model run, Labelbox will suggest some auto-generated slices. These auto-generated slices usually contain high-impact data.

Generate slices to surface high-impact data

Generate slices to surface high-impact data.

Below is the list of auto-generated slices available in Labelbox.

Auto-generated sliceDescription
True positivesSort data rows by the highest number of true positives
True negativesSort data rows by the highest number of true negatives
False positiveSort data rows by the highest number of false positives
False negativeSort data rows by the highest number of false negatives
Low precisionSort data rows by descending precision
Low recallSort data rows by descending recall
Low f1-scoreSort data rows by descending f1-score
Low confidenceSort data rows by descending confidence
Candidate mislabelsSurface candidate labeling mistakes by keeping images with false positives and sorting on descending confidence

Explore a slice

To explore a slice, go to the left side panel of the Model, click on Slices, and click on the slice name you want to explore.

The filter associated with the slice will show up instantly. All data rows matching the filter will show up as well.

Once you are done exploring a slice, you can:

  • Click on a different slice name, to explore a different slice.
  • Click on Datasets and a dataset name to explore a dataset.
  • Click on All data to explore your complete Catalog.

Update a slice

Over time, you might need to adjust the filters associated with a slice. Upon navigating to a slice, you can modify, add, or delete attributes of the filters. After you update the filters, save your changes by clicking on Update slice.

Labelbox will prompt you to choose between the following:

  1. Update the slice with the newly configured filters.
  2. Save as new a distinct slice with the newly configured filters.

Automate data curation using slices

Slices are dynamic, meaning that any data uploaded to Labelbox will automatically populate in any relevant slices. Therefore, the data rows in a slice may change over time in two possible ways:

  • New data rows appear in a slice: If you add new data rows to Catalog, they will appear in the slices whose filters they match. A data row can appear in many slices.
  • Existing data rows disappear from a slice: If a data row is deleted from Catalog, or if it no longer matches a slice filter, then it will not appear in the slice.

You can use slices to enable automated data curation workflows. For example, a possible workflow for surfacing high-impact data in Catalog may be:

  1. Set up filters in Catalog and assure that the filters are surfacing the intended data rows.
  2. Save the filter by creating a slice.
  3. Create a pipeline with cloud functions to automatically create new data rows in Labelbox when data lands in your storage buckets.
  4. Take action at any desired interval on high-impact data rows that automatically land in certain slices.

Python SDK

The ability to work with model slices via the Python SDK is coming soon.