This repository contains 690,269 analyzed faces obtained from Instagram posts. We have included the face attributes(age, gender, race, and etc.) and facial landmarks for each of these faces.

Instagram is a rich source for high-quality images, we believe those images can be tremendously helpful to the research community; however, Instagram restricts its API accesses to company only, and it builds many barricades for web crawlers. Thus, we decide to share the Instagram posts we retrieved. Further and more importantly, since detecting faces can cost numerous time and computing power, we share all those processed faces with the hope of saving time and efforts, facilitating the research for whom may concern.

We collected all faces from all acquired posts. We also recorded the timestamp, user_id, and post_id for each post, so that one can retrieve post from Instagram easily.

For each face, the included attributes are age, gender, race, and smiling. Along with face width and height, five facial landmarks are provided. They are eye_left, eye_right, nose, mouth_left, mouth_right, and center. The following figure illustrates the facial landmarks.

The following is one face example in JSON format. Note that one post may contain more than one face.

{
  	"timestamp": "1482223780",
  	"resp": [{
  		"attribute": {
  			"gender": {
  				"confidence": 77.270200000000003,
  				"value": "Male"
  			},
  			"age": {
  				"range": 6,
  				"value": 21
  			},
  			"race": {
  				"confidence": 54.615700000000004,
  				"value": "Asian"
  			},
  			"smiling": {
  				"value": 10.914
  			}
  		},
  		"face_id": "de7ff248bc19cb34e610070f21ee6210",
  		"tag": "",
  		"position": {
  			"eye_left": {
  				"y": 32.795499999999997,
  				"x": 54.653333000000003
  			},
  			"center": {
  				"y": 43.25,
  				"x": 57.416666999999997
  			},
  			"width": 29.833333,
  			"mouth_left": {
  				"y": 51.423499999999997,
  				"x": 48.733333000000002
  			},
  			"height": 29.833333,
  			"mouth_right": {
  				"y": 49.200000000000003,
  				"x": 58.359833000000002
  			},
  			"nose": {
  				"y": 42.856332999999999,
  				"x": 58.156999999999996
  			},
  			"eye_right": {
  				"y": 40.310833000000002,
  				"x": 66.170833000000002
  			}
  		}
  	}],
  	"uid_pid": "3040732185_1409317677457159168.jpg"
  }

For the above post, only one face is detected, and the uid is 3040732185, and pid is 1409317677457159168.

We can retrive Instagram post via pid; however, Instagram does not explicitly utilize pid to represent a link to a specific image, instead it utilizes shortcode. Generating shortcode from pid requires more work. Here is a detailed explaination about the logic behind how to construct shortcode. But we have included a complimentary converter (id2code.py) in this repository. Simply call

  python id2code.py 1467331451728886737 (or your own pid)

you will get the short code

  'BRdAi5cDvvR'

If you want the post directly, there is a function named generateUrl for help. Given pid, the full url points to the image post will be returned.

  'https://scontent-ord1-1.cdninstagram.com/t51.2885-15/s640x640/sh0.08/e35/17077495_255710568218080_4361264754778439680_n.jpg'

As mentioned above, first running

  python id2code.py 1467331451728886737

and, we can get the post's short-code, and the post's media address, where you can obtain the image directly.

   'BRdAi5cDvvR' # short-code
   'https://scontent-ort2-1.cdninstagram.com/t51.2885 15/s640x640/sh0.08/e35/17077495_255710568218080_4361264754778439680_n.jpg' # image address

Then, with the image address, you can either view the image through the browser, or use any code to download it to your disk, and etc. In this case:

user face is blurred for privacy consideration.