CODEWITHSUNDEEP
javaibm-watsonvisual-recognition
ZaCH
ZaCH

Reputation: 31

IBM Watson Visual Recognition: Results format in Java

I'd like to know if is there a possibility to change the format of result returned bi Watson Visual Recognition API.

For example:

Instead of having this:

wasp-nest: 0.98, bird-nest: 0.9, hornet-nest: 0.95

Get this:

wasp-nest: 0.98, bird-nest: 0.0, hornet-nest: 0.02

So the sum of classes results would be 1 (100%)

Upvotes: 0

Views: 92

Answers (1)

chughts
chughts

Reputation: 4735

That is not how the machine learning algorithm that the Watson Visual Recognition service uses works. It is checking against all the classifiers that you have requested, which by default is all the inbuilt classifiers, and will return the confidence level that the image matches against each classifier individually.

There is no mutual exclusion across classifiers, i.e. if A then it can't be B. Instead its confidence level for a match on A and a confidence level for a match on B. The result is Xa% confidence for A and Xb% confidence for B.

As a wasp's nest will have some similarity to a hornet's nest, you would expect the classification match to score high on both. It should record low on say a brick house.

Upvotes: 3

Related Questions