Eye/Facial Coordinates in JSON?

[rgvxlens]

Hi there, is it possible to print/output the Facial Coordinates/Pupil Localization pointers in the JSON file with -json itself? Or any work around for the same?

or any other way to print the exact coordinates of the pupils or the other facial features?

[esimov]

Currently only the face coordinates could be exported into a JSON file, but this can be extended to support the pupil and facial coordinates.

[rgvxlens]

Thanks for the your quick response!

But can this still be done if I create an API for the same or a Python program? Just need to output the coordinates somewhere.

[esimov]

You mean would you like such a feature to be included into the library? If that's what you want I can take care of it.

[rgvxlens]

Would be awesome! or maybe a way to just get specific features in the JSON? Both would be helpful, or whatever is possible according to you.

[esimov]

@rgvxlens i've extended the json marshalizer to include the pupils/eyes localization and landmark points detection coordinates. Currently the json export is looking something like this:

[{
    "face": {
        "x": 129,
        "y": 131,
        "size": 158
    },
    "eyes": [{
        "x": 116,
        "y": 107,
        "size": 12
    }, {
        "x": 114,
        "y": 162,
        "size": 13
    }],
    "landmark_points": [{
        "x": 103,
        "y": 185,
        "size": 19
    }, {
        "x": 91,
        "y": 168,
        "size": 19
    }, {
        "x": 103,
        "y": 147,
        "size": 19
    }, {
        "x": 115,
        "y": 152,
        "size": 19
    }, {
        "x": 113,
        "y": 175,
        "size": 19
    }, {
        "x": 153,
        "y": 134,
        "size": 19
    }, {
        "x": 184,
        "y": 113,
        "size": 19
    }, {
        "x": 192,
        "y": 133,
        "size": 19
    }, {
        "x": 179,
        "y": 134,
        "size": 19
    }, {
        "x": 183,
        "y": 160,
        "size": 19
    }]
}]

You might ignore the size in eyes and landmark_points since they are not really relevant.

[rgvxlens]

Thanks for this! What does the size in the face denote?

[esimov]

The dimension of the detected face region.

[rgvxlens]

Alright, i'll give it a look and report back. Thanks a lot and kudos to you for this amazing project.

[rgvxlens]

Hey! I had a few questions, I hope you don't mind answering them:

1. Why is my output always like this?

   [{"Min":{"X":33,"Y":81},"Max":{"X":243,"Y":243}}]

The rest of the JSON file is empty other than the string above.

This is the command that I'm running: pigo -in /home/rgvx/githubstuff/RedEyeRemover/test.jpeg -out output.jpg -cf /home/rgvx/githubstuff/pigo/cascade/facefinder -mark -pl -plc /home/rgvx/githubstuff/pigo/cascade/puploc -json

Is it a problem with the command I'm running?

2. What is the format of the "size" value? Is it in Pixels?

Thanks!

Edit: Figured that I was on the wrong version of Pigo (the one that I got from the AUR). Ran the latest version and now I get this:

[{"face":{"x":422,"y":336,"size":483},"eyes":[{"x":378,"y":267,"size":39},{"x":380,"y":432,"size":40}]}]

`

[esimov]

Every value in the json file is in pixel format. The size value as I mentioned earlier in the thread represents the detected face rectangle size, but since this might be relevant only for the detected face you can ignore on pupil/eyes localization and facial landmark detection, but it's up to you.

[rgvxlens]

Oh, thanks! Got it, I'm now looking into calculating/detecting the radius of the eyes with the coordinates that I've obtained. Do you have any idea about the same?

[esimov]

You don't need to calculate the radius of the eyes, you only need to multiply the scale value with a percentage, 0.4 should give you an approximate correct result. (https://twitter.com/simo_endre/status/1171002326812962816)

[rgvxlens]

That's exactly what I'm looking for. Thank you so much mate! What exactly is that value? Can you give me a bit of a background on that? Just so that I have some background info on that.

Edit: Oh nevermind, I think the value arises from Hough Circles. I'll do a deeper research into them. Thanks a ton once again! :)

[esimov]

@rgvxlens i'm really glad that it worked for you! ;)