0

Per this link:

https://cloud.google.com/vision/docs/internet-detection

"Here is the Python command with the passed-in Web URL of the car image, followed by console output. Note that a relevancy score is added after the listed entities. Note that scores are not normalized or comparable across different image queries."

If scores are not normalized, what do they indicate? For the highest score out of all the listed results from a single image query, how do you measure the importance of that result to your query? How do you get any useful information beyond the relative rank of results?

More specifically though, does anyone know how the values are calculated?

Thanks

Josh Flori
  • 104
  • 11

1 Answers1

0

If I am not wrong, your question is about WebEntity score. This is from the API Reference for WebEntity:

Overall relevancy score for the entity. Not normalized and not comparable across different image queries.

I am not sure how the scores are calculated behind the scene. But I think for a single image query you can still normalize the scores of the 5 entities from [0, maxScore] to [0-1] to get sort of a confidence score.

Taking the same example of the link:

The original scores for (Volkswagen Beetle, Volkswagen,Volkswagen Type 2, Van, Car) = (5.35028934479, 1.43998003006, 0.828279972076, 0.75271999836, 0.690039992332)

Normalized scores = (1.00000000000, 0.269140589838, 0.154810313742, 0.14068771796, 0.128972462584)

Xiaoxia Lin
  • 736
  • 6
  • 16
  • Yes sorry, i was referring to the WebEntity score. Sure, that makes sense that you can normalize them for each result, but the real big question is whether a normalized score is comparable to another normalized score. I just don't think so. I love the api otherwise though. – Josh Flori Apr 02 '18 at 17:56
  • I don't think so either if another normalized score is from a different image query. – Xiaoxia Lin Apr 05 '18 at 11:12