Uncertainty score

The API returns a confidence score of the cutout through the HTTP header x-uncertainty-score.

The uncertainty score is a number between 0 and 1:

  • 0 means the model is very confident that the cutout is accurate

  • 1 means the model is unsure.

For instance, shoes on a shoe box might give a higher uncertainty score as the model is unsure what to segment (shoes, box or both):

High uncertainty score: between 0.7 and 1

Low uncertainty score: between 0 and 0.3

In some cases, we might not be able to able to return an uncertainty score (e.g. if an image contains humans), in that case the value -1 will be returned.

Last updated

Was this helpful?