Custom Vision API

Question

I want to count number of people on below image,
I believe it could be possible with Microsoft vision api or custom vision api but right now I'm not able to login on microsoft cognitive service.
Can anyone please confirm, Is it possible or not with MS cognitive services ?

I don't think it's possible. You can use Face API to count the number of faces. — Maria Ines Parnisari, May 13 '17 at 02:43
Yes, you are right. do you know any other workaround to count number of objects — Pankaj Rawat, May 13 '17 at 07:08

score 1 · Answer 1 · answered Jun 05 '17 at 20:24

1

Counting objects within images is currently not supported for any of the Vision API's provided through Microsoft Cognitive Services.

answered Jun 05 '17 at 20:24

LeviJ3ans

296
2
3

Is this still the case? – Meneghino May 15 '18 at 13:10

score 1 · Answer 2 · answered Aug 22 '17 at 23:51

You can test it on your own using this url - https://azure.microsoft.com/en-us/services/cognitive-services/computer-vision/ -> Go to 'Recognize celebrities and landmarks' This is what is the output

{
"categories": [
{
  "name": "plant_tree",
  "score": 0.45703125,
  "detail": null
}
],
"adult": null,
"tags": [
{
  "name": "grass",
  "confidence": 0.99937587976455688
},
{
  "name": "field",
  "confidence": 0.94717073440551758
},
{
  "name": "people",
  "confidence": 0.65122932195663452
},
{
  "name": "grassy",
  "confidence": 0.546949565410614
},
{
  "name": "lush",
  "confidence": 0.11115188896656036
}
],
"description": {
"tags": [
  "grass",
  "kite",
  "field",
  "flying",
  "people",
  "grassy",
  "large",
  "soccer",
  "group",
  "man",
  "green",
  "walking",
  "dog",
  "playing",
  "many",
  "hill",
  "woman",
  "blue",
  "standing",
  "crowd",
  "ball"
 ],
 "captions": [
  {
    "text": "a group of people flying kites in a field",
    "confidence": 0.7429714961690741
  }
 ]
 },


 "requestId": "ba0592e6-1449-450e-a00f-a551ed0da4a8",
  "metadata": {
    "width": 160,
    "height": 120,
    "format": "Jpeg"
  },



"faces": [],
  "color": {
    "dominantColorForeground": "Green",
    "dominantColorBackground": "Green",
    "dominantColors": [
      "Green"
    ],
    "accentColor": "718E39",
    "isBWImg": false
  },
  "imageType": {
    "clipArtType": 0,
    "lineDrawingType": 0
  }
}

It identifies people with a confidence level of 0.65%

I tried with a more clear image and got the count of people in the form of "faces" enter image description here

score 1 · Answer 3 · answered Apr 11 '19 at 00:15

Custom Vision API and Computer Vision API both supports object detection functionality now.

If you have an existing dataset or you can label a dataset, then you can use Custom Vision to train a model first and then detect similar persons in unseen images. If you are ok with general person detector, which does not need training, then you can use Computer Vision API.

Custom Vision API

3 Answers3