1

Is there a way to get Watson Visual Recognition to return the location of the classified content (car, tree, etc.) when using image classification? This capability exists in the face recognition service and would be invaluable in general image classification.

The current documentation for image classification has no information on this topic.

Bob Dill
  • 1,000
  • 5
  • 13
  • The current Watson Visual Recognition (classify images for face recognition) doesn't have this feature too! When I call `/v3/classify` on an image with multiple faces, I only get the class name and the score. There is no information about the location of the recognized face in the picture, while you get location information for face detection `/v3/detect_faces` Right? – Peshmerge Apr 16 '18 at 14:27
  • @Peshmerge you are correct. Image classification (/v3/classify) is comparing the provided image with the classes of images in your classifier. It is not (necessarily) doing object level recognition as part of the /v3/classify/ request. I do not find any documentation in the current Visual Recognition documentation where object location information is provided other than in the detect_faces API – Bob Dill Apr 20 '18 at 16:59

1 Answers1

1

We recognize the value of object localization, however this function is not provided by the visual recognition service. One approach some users have taken, is to segment their image first and then submit the segments of interest to the service.

Matt Hill
  • 1,081
  • 6
  • 4
  • I hope that IBM will choose to offer this capability in the future. Visual Recognition already has this capability, as shown in the Detect Faces service. Thanks for the answer. – Bob Dill Nov 09 '16 at 14:59