Google Cloud Vision API test: update

Index, feed.

Previously, Google's Vision API would find dangerous weapons in the following pictures, with the confidence scores reported:

Gabriel Nadeau-Dubois, Jeanne Reynolds

Today, in late July 2017, those are the labels we get from the same picture:

{
  "labelAnnotations": [
    {
      "mid": "/m/0k4j",
      "description": "car",
      "score": 0.87314415
    },
    {
      "mid": "/m/03qtwd",
      "description": "crowd",
      "score": 0.83961844
    },
    {
      "mid": "/m/07yv9",
      "description": "vehicle",
      "score": 0.7199295
    },
    {
      "mid": "/m/081pkj",
      "description": "event",
      "score": 0.67535704
    },
    {
      "mid": "/m/05r655",
      "description": "girl",
      "score": 0.61985064
    }
  ]
}

In table form:

Compared to the data from a year ago:

• the most egregious label, "guided missile destroyer", is gone;
• we have a new relevant label "crowd", and correct but perhaps less exciting labels "event" and "girl";
• but Google still insists incorrectly on "vehicle", the only label common to the two responses, and even adds "car".

www.kurokatta.org

David Haguenauer <www@kurokatta.org>
$Id: google-vision-crowd.malblog,v 1.3 2017/08/08 23:31:01 homer Exp $