Review assets

Every asset in a dataset has a review status. Use it to track what still needs work, record review decisions, and choose which data goes into training.

Review statuses

Status	Meaning
To annotate	The asset has no annotations yet, or was sent back for annotation.
Pending review	The asset has annotations that nobody has reviewed yet.
Approved	A reviewer accepted the asset.
Rejected	A reviewer rejected the asset.

New assets start as To annotate. Some changes happen automatically:

• When an asset in To annotate gets its first annotation, it moves to Pending review.
• When Magic Annotate or Magic Select adds annotations to an Approved or Rejected asset, it moves back to Pending review.

Approved and Rejected are always set by a person. Document datasets use the same statuses for whole documents. See Documents and OCR.

Find assets to review

• In Datasets, select the n to review link on a dataset to open it filtered to Pending review.
• On the dataset page, select Filter and choose one or more statuses under Review status. You can combine them with the Annotations and Labels filters.

The filters carry into the editor, so you can review the matching images one after the other.

Review in the editor

1. Open the first image to review, for example from the filtered asset grid.
2. Check every annotation. Fix, relabel, or delete what is wrong. See Annotate images.
3. In the top bar, select a status: To annotate, Pending review, Approved, or Rejected. You can also press 1, 2, 3, or 4.
4. Press → or select Next asset to move to the next image.

Change the status of many assets

1. On the dataset page, select assets: use the selection circle on each tile in the grid view, the checkboxes in the list view, or Select all.
2. In the bar that appears at the bottom, select Approve, or open Change status and choose Send back to annotation, Mark as pending review, Approve, or Reject.

You can select up to 100 assets at a time. Changing the filters clears the selection. The same bar also has Delete, which permanently deletes the selected assets and their annotations.

Why approval matters

• Dataset versions capture the review status of every asset at the moment you create them. Later changes do not affect an existing version. See Dataset versions.
• Model training uses only the approved images of a ready image dataset version. Approved images without annotations are used as negative examples. See Train a model.
• Knowledge bases can index only approved documents, with the Approved documents only option. See Knowledge bases.

Permissions

Changing a review status needs edit access to the dataset. With view-only access, you can see the statuses and filter by them. See Datasets.

Next steps

• Dataset versions
• Train a model
• Export datasets