Skip to main content
Visual Layer reduces dataset redundancy by highlighting the most visually distinct samples in your dataset. This workflow:
  • Reduces redundancy before model training or export.
  • Creates compact datasets that preserve diversity.
  • Reveals long tail content often hidden behind common or repetitive examples.
Visual Layer assigns each image or object a uniqueness score, then filters out near-duplicates according to a user-defined threshold.

How It Works

1

Uniqueness Scoring

Visual Layer assigns every image or object a uniqueness score between 0 and 1:
  • 0 = highly redundant.
  • 1 = highly unique.
2

Filter Application

Visual Layer retains only samples with the highest uniqueness scores when you apply the filter
3

Query Integration

The action is automatically applied as an Exclude rule in the Query Panel

How to Apply the Filter

This workflow lets you focus on the most varied content in your dataset—ideal for debugging, QA, or preparing smaller benchmark sets.
  1. Navigate to the Dataset Exploration View.
  2. Click the Select Uniques button (located next to the search bar).
  3. Adjust the threshold slider to your preferred level of uniqueness.
  4. The filter will appear in the Query Panel under the Exclude section.

Uniqueness in Exported Data

If you export data with this filter applied, the retained items will include their uniqueness score in the JSON payload:
"uniqueness_score": 0.9285714285714288
I