​
What Are Dark Images and Objects?

Dark images are underexposed visuals that lack sufficient brightness, making it difficult to interpret or extract visual details. They often appear dim, shadowed, or low-contrast—reducing both human and model-level interpretability.

​
Common Causes of Darkness in Visual Data

  • Poor lighting conditions: Capturing in low-light environments or scenes without adequate illumination leads to dark outputs.
  • Camera settings: Improper settings like fast shutter speed, low ISO, or narrow aperture can result in underexposed images.
  • Scene characteristics: Some subjects—like night scenes, shaded objects, or predominantly dark materials—naturally produce low-light imagery.

​
Why It Matters

ProblemImpact
Visual ambiguityAnnotators may miss important elements due to lack of detail or poor visibility.
Model degradationObject detection and segmentation performance can drop significantly on low-exposure inputs.
Poor user experienceIn public-facing applications, dark visuals can appear unpolished or unprofessional.

​
How to Detect and Remove Dark Images

  • Detect Dark Images:
    Go to “Add Filter” → “Quality Issue” → select “Dark” → choose “IS” as the logic operator → set the desired confidence threshold (default is 0.5).
    Export the dark images using Export → “Matching the applied filter.”
  • Remove Dark Images:
    Go to “Add Filter” → “Quality Issue” → select “Dark” → choose “IS NOT” as the logic operator → set the desired confidence threshold (default is 0.5).
    Export the images without dark issues using Export → “Matching the applied filter.”
By detecting and resolving dark visuals early, you can improve dataset quality, annotation accuracy, and model performance—especially in environments where lighting is a critical factor.