search

BSI-OFIQ / OFIQ-Project

Open Source Facial Image Quality
Other
38 stars 12 forks source link

Rotate 180 degrees is not clearly detected #32

Open tadam98s opened 3 months ago

tadam98s commented 3 months ago

When a face is rotated 180 degrees I do not see a clear detetion. +90/-90 are detected at -80/+84 which is fine. Can you advise?

Best, Mickey 2021-MickeyCohen 2021-MickeyCohen-90 2021-MickeyCohen-180 2021-MickeyCohen-m90 NoFace table.csv table.xlsx

bbtams commented 3 months ago

Hi Mickey, yes it may be true that the HeadPose-Estimation algorithm specified in DIS 29794-5 does not detect such extrem pose deviations. But this is not an issue of the OFIQ code here.

tadam98s commented 3 months ago

Hi, not sure that a 180 degrees rotated face would be recognized by the face engines. I would have expected some rejection on such a face.

christophbusch commented 3 months ago

Hi Mickey,

thanks for evaluating OFIQ 1.0 and ISO/IEC DIS 29794-5. Your observation is valid and this was not covered yet.

I suggest that for now we add to the definition of a 'face image' in Clause 3.4 and to the OFIQ documentation the following sentence: "A face image anticipates a capture that is conformant to policies regulating acquisition of biometric samples. A capture attempt, which generates an image that is rotated 90 degrees or more (or even upside down) is not conformant to such policies."

In OFIQ 2.0 we can address then such extreme roll angles and detecting images that are upside down.

best regards Christoph

tadam98s commented 3 months ago 
body p { margin-bottom: 0cm; margin-top: 0pt; } 

Hi,

I also have many misses of multiple faces in
    the image. Many of those images get high score and also noted as
    single face.
Busy background is not well indicated.

I have incorporated the latest version
    in-series with incoming images that people upload from home. We
    have a new regulation that you can ask to issue a biometric
    document based on biometrics already in the depository provided
    you pass 1:1 with an uploaded self image (which is not included
    in the document. only used for 1:1).

I have already tested 7,700 images and more to
    follow.

  I would appreciate an "eye-glasses
      detector".
  Also, there is no clear indication about
      minimal passe-par·tout and many of the faces where the
      forehead touches the top border pass without comment.
  Eye (pupils) distance

I think I am trying two things is parallel:
1. Is the image per ISO/IEC 79794-5 so that we
    can guide the officer/data-subject to re-take a better image.
2. Consolidation of image contents and a
    statistical indication of success in 1:M recognition.

While we generally agree that (1) is not
    necessary to achieve (2), especially with the modern AI based
    recognition systems, focusing only on (2) may place the effort
    OUTSIDE ISO/IEC 79794-5.

Contining the study.

  body p { margin-bottom: 0cm; margin-top: 0pt; } 

  Best regards,
   Mickey Cohen
   Shanit Ltd.
   CEO
   POB 23410, Jerusalem 9123302, Israel
    Schedule a meeting
    (M) +972-54-758-6312
   Skype: tadam_98 
   ***@***.***

On 19/05/2024 9:51 AM, Christoph Busch
  wrote:

  Hi Mickey,
  thanks for evaluating OFIQ 1.0 and ISO/IEC DIS
    29794-5.
    Your observation is valid and this was not covered yet.
  I suggest that for now we add to the definition of a
    'face image' in
    Clause 3.4 and to the OFIQ documentation the following sentence:
    "A face image anticipates a capture that is conformant to
    policies
    regulating acquisition of biometric samples. A capture attempt,
    which
    generates an image that is rotated 90 degrees or more (or even
    upside
    down) is not conformant to such policies."
  In OFIQ 2.0 we can address then such extreme roll
    angles and detecting
    images that are upside down.
  best regards
    Christoph
  —hj
    Reply to this email directly, view it on GitHub, or unsubscribe.
    You are receiving this because you authored the thread.Message
      ID: ***@***.***>
  [

{ @.": "http://schema.org", @.": "EmailMessage", "potentialAction": { @.": "ViewAction", "target": "https://github.com/BSI-OFIQ/OFIQ-Project/issues/32#issuecomment-2119124650", "url": "https://github.com/BSI-OFIQ/OFIQ-Project/issues/32#issuecomment-2119124650", "name": "View Issue" }, "description": "View this Issue on GitHub", "publisher": { @.": "Organization", "name": "GitHub", "url": "https://github.com" } } ]

--------------q81ajuaePQ4xEti6gDRdXgyR Content-Type: image/jpeg; name="7k6RkBB5dGqFfsQr.jpeg" Content-Disposition: inline; filename="7k6RkBB5dGqFfsQr.jpeg" Content-Id: Content-Transfer-Encoding: base64

/9j/4AAQSkZJRgABAQEAYABgAAD/2wBDAAIBAQIBAQICAgICAgICAwUDAwMDAwYEBAMFBwYH BwcGBwcICQsJCAgKCAcHCg0KCgsMDAwMBwkODw0MDgsMDAz/2wBDAQICAgMDAwYDAwYMCAcI DAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAz/wAAR CAAMAAwDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAA AgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkK FhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWG h4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl 5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREA AgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYk NOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOE hYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk 5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD79PxatfG3wvj1jSxqOpeJLPUJE1tUYzP5 UpBgmEYy20sShYA4f72NyVgaX+1xqvhSOWzs9UvtPVJW82FZWj2SDhgV7EYwQeeK+afi5daj +zN+0/4lsvB+taxpcnhjVJ7awu1uSLhEDYwzAAMMcEEYIHIPNfU37Gn7Jeg/tvfDTU/iF8Qt W8Sal4o1PWJobm5guo7dZBHHEF+RY9o444AHA4r+NsDXxGPxX1fBy9niYe7J3tCSjo5cy97m k7XTi7u8m9bH8K5dxBmGZ4v6ngpezxdO8ZO7UJKGjkpJc/M3a6cWm7yck3Y//9k= --------------q81ajuaePQ4xEti6gDRdXgyR Content-Type: image/jpeg; name="3vcu7hyDy3jRtUu0.jpeg" Content-Disposition: inline; filename="3vcu7hyDy3jRtUu0.jpeg" Content-Id: Content-Transfer-Encoding: base64

/9j/4AAQSkZJRgABAQEAYABgAAD/2wBDAAIBAQIBAQICAgICAgICAwUDAwMDAwYEBAMFBwYH BwcGBwcICQsJCAgKCAcHCg0KCgsMDAwMBwkODw0MDgsMDAz/2wBDAQICAgMDAwYDAwYMCAcI DAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAz/wAAR CAAMAAwDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAA AgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkK FhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWG h4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl 5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREA AgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYk NOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOE hYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk 5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD7S+Fn7YPh/wCKVt8QLe6gjnuvAfgTWPF0 0i3TBp3smjwpVDny9sqhsDcWBIYZAr55T/go6/iK+vLjRF1nTdPMiqsCyNtVxGm8gkglS24j PIGASSM15PBrmqfs6/Enx9Y+E9QWxh1d77w/qDXGnWV619YNMyyW7+fC48uQKvmIoCvtXcDt XHo37Hf7J/hv4m/DTUL6+utYhlg1SS2VbeWMKVEMLZO6NjnLHv6V/Qb4NwuGjKvU1i7cuuvn dbLXazfyPyZcRV60o0oaSV7/AKWe/wCB/9k= --------------q81ajuaePQ4xEti6gDRdXgyR Content-Type: image/jpeg; name="qDb6oKabpNfmKY5T.jpeg" Content-Disposition: inline; filename="qDb6oKabpNfmKY5T.jpeg" Content-Id: Content-Transfer-Encoding: base64

/9j/4AAQSkZJRgABAQEAYABgAAD/2wBDAAIBAQIBAQICAgICAgICAwUDAwMDAwYEBAMFBwYH BwcGBwcICQsJCAgKCAcHCg0KCgsMDAwMBwkODw0MDgsMDAz/2wBDAQICAgMDAwYDAwYMCAcI DAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAz/wAAR CAAMAAwDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAA AgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkK FhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWG h4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl 5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREA AgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYk NOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOE hYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk 5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD78/Y9/Y/0fxv4Q1jxP4o1jwv4iuPEOhy6 hDDf3H+maVc5z9onySdmWJLMcEFSQc8eAfFfSvEn7LHjGbwvo3jq7urUxpeNPomoSw2kzuME jY+GYBQCfVcdqTSvitrH7LPxJ8daP4Xkha1uZ7rRJmvIhNJLbJK6bSRgAsAMkAHI4xXd/sif sieG/j58Nr7WNYvtctbq21OSzVbKaJIyixROCQ0bHOZD3xjHFfxHmmNwGKyrCYPL8P7KvHnv PnfR+9qtZcz1974do6H9p5fhMbhMzxeNzHEe1oS5LQ5I9V7uj0jyLT3W+bd6n//Z

--------------q81ajuaePQ4xEti6gDRdXgyR--

tadam98s commented 3 months ago 
Hi,

Out of 120 heads rotated +90 or -90 it
    detected only 6 as rotated.
Something with the rotation detector is not
    working as expected.

  body p { margin-bottom: 0cm; margin-top: 0pt; } 

  Best regards,
   Mickey Cohen
   Shanit Ltd.
   CEO
   POB 23410, Jerusalem 9123302, Israel
    Schedule a meeting
    (M) +972-54-758-6312
   Skype: tadam_98 
   ***@***.***

On 19/05/2024 10:03 AM, Mickey Cohen
  wrote:

  body p { margin-bottom: 0cm; margin-top: 0pt; }
  Hi,

  I also have many misses of multiple faces in
      the image. Many of those images get high score and also noted
      as single face.
  Busy background is not well indicated.

  I have incorporated the latest version
      in-series with incoming images that people upload from home.
      We have a new regulation that you can ask to issue a biometric
      document based on biometrics already in the depository
      provided you pass 1:1 with an uploaded self image (which is
      not included in the document. only used for 1:1).

  I have already tested 7,700 images and more
      to follow.

    I would appreciate an "eye-glasses
        detector".
    Also, there is no clear indication about
        minimal passe-par·tout and many of the faces where the
        forehead touches the top border pass without comment.
    Eye (pupils) distance

  I think I am trying two things is parallel:
  1. Is the image per ISO/IEC 79794-5 so that
      we can guide the officer/data-subject to re-take a better
      image.
  2. Consolidation of image contents and a
      statistical indication of success in 1:M recognition.

  While we generally agree that (1) is not
      necessary to achieve (2), especially with the modern AI based
      recognition systems, focusing only on (2) may place the effort
      OUTSIDE ISO/IEC 79794-5.

  Contining the study.

    body p { margin-bottom: 0cm; margin-top: 0pt; }

    Best regards,
     Mickey Cohen
     Shanit Ltd.
     CEO
     POB 23410, Jerusalem 9123302, Israel
      Schedule
        a meeting
      (M)
      +972-54-758-6312
     Skype: tadam_98

     ***@***.***

  On 19/05/2024 9:51 AM, Christoph
    Busch wrote:

    Hi Mickey,
    thanks for evaluating OFIQ 1.0 and ISO/IEC DIS
      29794-5.
      Your observation is valid and this was not covered yet.
    I suggest that for now we add to the definition of
      a 'face image' in
      Clause 3.4 and to the OFIQ documentation the following
      sentence:
      "A face image anticipates a capture that is conformant to
      policies
      regulating acquisition of biometric samples. A capture
      attempt, which
      generates an image that is rotated 90 degrees or more (or even
      upside
      down) is not conformant to such policies."
    In OFIQ 2.0 we can address then such extreme roll
      angles and detecting
      images that are upside down.
    best regards
      Christoph
    —hj
      Reply to this email directly, view it on GitHub, or unsubscribe.
      You are receiving this because you authored the thread.Message
        ID: ***@***.***>
    [

{ @.": "http://schema.org", @.": "EmailMessage", "potentialAction": { @.": "ViewAction", "target": "https://github.com/BSI-OFIQ/OFIQ-Project/issues/32#issuecomment-2119124650", "url": "https://github.com/BSI-OFIQ/OFIQ-Project/issues/32#issuecomment-2119124650", "name": "View Issue" }, "description": "View this Issue on GitHub", "publisher": { @.": "Organization", "name": "GitHub", "url": "https://github.com" } } ]

--------------GwlPkhZ0dHd5giKQOtFyACWy Content-Type: image/jpeg; name="uKmvt5FONLRLbgyO.jpeg" Content-Disposition: inline; filename="uKmvt5FONLRLbgyO.jpeg" Content-Id: Content-Transfer-Encoding: base64

/9j/4AAQSkZJRgABAQEAYABgAAD/2wBDAAIBAQIBAQICAgICAgICAwUDAwMDAwYEBAMFBwYH BwcGBwcICQsJCAgKCAcHCg0KCgsMDAwMBwkODw0MDgsMDAz/2wBDAQICAgMDAwYDAwYMCAcI DAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAz/wAAR CAAMAAwDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAA AgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkK FhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWG h4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl 5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREA AgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYk NOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOE hYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk 5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD79PxatfG3wvj1jSxqOpeJLPUJE1tUYzP5 UpBgmEYy20sShYA4f72NyVgaX+1xqvhSOWzs9UvtPVJW82FZWj2SDhgV7EYwQeeK+afi5daj +zN+0/4lsvB+taxpcnhjVJ7awu1uSLhEDYwzAAMMcEEYIHIPNfU37Gn7Jeg/tvfDTU/iF8Qt W8Sal4o1PWJobm5guo7dZBHHEF+RY9o444AHA4r+NsDXxGPxX1fBy9niYe7J3tCSjo5cy97m k7XTi7u8m9bH8K5dxBmGZ4v6ngpezxdO8ZO7UJKGjkpJc/M3a6cWm7yck3Y//9k= --------------GwlPkhZ0dHd5giKQOtFyACWy Content-Type: image/jpeg; name="5EydazwoKXd1vD9l.jpeg" Content-Disposition: inline; filename="5EydazwoKXd1vD9l.jpeg" Content-Id: Content-Transfer-Encoding: base64

/9j/4AAQSkZJRgABAQEAYABgAAD/2wBDAAIBAQIBAQICAgICAgICAwUDAwMDAwYEBAMFBwYH BwcGBwcICQsJCAgKCAcHCg0KCgsMDAwMBwkODw0MDgsMDAz/2wBDAQICAgMDAwYDAwYMCAcI DAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAz/wAAR CAAMAAwDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAA AgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkK FhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWG h4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl 5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREA AgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYk NOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOE hYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk 5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD7S+Fn7YPh/wCKVt8QLe6gjnuvAfgTWPF0 0i3TBp3smjwpVDny9sqhsDcWBIYZAr55T/go6/iK+vLjRF1nTdPMiqsCyNtVxGm8gkglS24j PIGASSM15PBrmqfs6/Enx9Y+E9QWxh1d77w/qDXGnWV619YNMyyW7+fC48uQKvmIoCvtXcDt XHo37Hf7J/hv4m/DTUL6+utYhlg1SS2VbeWMKVEMLZO6NjnLHv6V/Qb4NwuGjKvU1i7cuuvn dbLXazfyPyZcRV60o0oaSV7/AKWe/wCB/9k= --------------GwlPkhZ0dHd5giKQOtFyACWy Content-Type: image/jpeg; name="sz2RiCkYbcm91An4.jpeg" Content-Disposition: inline; filename="sz2RiCkYbcm91An4.jpeg" Content-Id: Content-Transfer-Encoding: base64

/9j/4AAQSkZJRgABAQEAYABgAAD/2wBDAAIBAQIBAQICAgICAgICAwUDAwMDAwYEBAMFBwYH BwcGBwcICQsJCAgKCAcHCg0KCgsMDAwMBwkODw0MDgsMDAz/2wBDAQICAgMDAwYDAwYMCAcI DAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAz/wAAR CAAMAAwDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAA AgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkK FhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWG h4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl 5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREA AgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYk NOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOE hYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk 5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD78/Y9/Y/0fxv4Q1jxP4o1jwv4iuPEOhy6 hDDf3H+maVc5z9onySdmWJLMcEFSQc8eAfFfSvEn7LHjGbwvo3jq7urUxpeNPomoSw2kzuME jY+GYBQCfVcdqTSvitrH7LPxJ8daP4Xkha1uZ7rRJmvIhNJLbJK6bSRgAsAMkAHI4xXd/sif sieG/j58Nr7WNYvtctbq21OSzVbKaJIyixROCQ0bHOZD3xjHFfxHmmNwGKyrCYPL8P7KvHnv PnfR+9qtZcz1974do6H9p5fhMbhMzxeNzHEe1oS5LQ5I9V7uj0jyLT3W+bd6n//Z --------------GwlPkhZ0dHd5giKQOtFyACWy Content-Type: image/jpeg; name="7k6RkBB5dGqFfsQr.jpeg" Content-Disposition: inline; filename="7k6RkBB5dGqFfsQr.jpeg" Content-Id: Content-Transfer-Encoding: base64

/9j/4AAQSkZJRgABAQEAYABgAAD/2wBDAAIBAQIBAQICAgICAgICAwUDAwMDAwYEBAMFBwYH BwcGBwcICQsJCAgKCAcHCg0KCgsMDAwMBwkODw0MDgsMDAz/2wBDAQICAgMDAwYDAwYMCAcI DAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAz/wAAR CAAMAAwDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAA AgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkK FhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWG h4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl 5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREA AgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYk NOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOE hYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk 5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD79PxatfG3wvj1jSxqOpeJLPUJE1tUYzP5 UpBgmEYy20sShYA4f72NyVgaX+1xqvhSOWzs9UvtPVJW82FZWj2SDhgV7EYwQeeK+afi5daj +zN+0/4lsvB+taxpcnhjVJ7awu1uSLhEDYwzAAMMcEEYIHIPNfU37Gn7Jeg/tvfDTU/iF8Qt W8Sal4o1PWJobm5guo7dZBHHEF+RY9o444AHA4r+NsDXxGPxX1fBy9niYe7J3tCSjo5cy97m k7XTi7u8m9bH8K5dxBmGZ4v6ngpezxdO8ZO7UJKGjkpJc/M3a6cWm7yck3Y//9k= --------------GwlPkhZ0dHd5giKQOtFyACWy Content-Type: image/jpeg; name="3vcu7hyDy3jRtUu0.jpeg" Content-Disposition: inline; filename="3vcu7hyDy3jRtUu0.jpeg" Content-Id: Content-Transfer-Encoding: base64

/9j/4AAQSkZJRgABAQEAYABgAAD/2wBDAAIBAQIBAQICAgICAgICAwUDAwMDAwYEBAMFBwYH BwcGBwcICQsJCAgKCAcHCg0KCgsMDAwMBwkODw0MDgsMDAz/2wBDAQICAgMDAwYDAwYMCAcI DAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAz/wAAR CAAMAAwDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAA AgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkK FhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWG h4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl 5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREA AgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYk NOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOE hYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk 5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD7S+Fn7YPh/wCKVt8QLe6gjnuvAfgTWPF0 0i3TBp3smjwpVDny9sqhsDcWBIYZAr55T/go6/iK+vLjRF1nTdPMiqsCyNtVxGm8gkglS24j PIGASSM15PBrmqfs6/Enx9Y+E9QWxh1d77w/qDXGnWV619YNMyyW7+fC48uQKvmIoCvtXcDt XHo37Hf7J/hv4m/DTUL6+utYhlg1SS2VbeWMKVEMLZO6NjnLHv6V/Qb4NwuGjKvU1i7cuuvn dbLXazfyPyZcRV60o0oaSV7/AKWe/wCB/9k= --------------GwlPkhZ0dHd5giKQOtFyACWy Content-Type: image/jpeg; name="qDb6oKabpNfmKY5T.jpeg" Content-Disposition: inline; filename="qDb6oKabpNfmKY5T.jpeg" Content-Id: Content-Transfer-Encoding: base64

/9j/4AAQSkZJRgABAQEAYABgAAD/2wBDAAIBAQIBAQICAgICAgICAwUDAwMDAwYEBAMFBwYH BwcGBwcICQsJCAgKCAcHCg0KCgsMDAwMBwkODw0MDgsMDAz/2wBDAQICAgMDAwYDAwYMCAcI DAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAz/wAAR CAAMAAwDASIAAhEBAxEB/8QAHwAAAQUBAQEBAQEAAAAAAAAAAAECAwQFBgcICQoL/8QAtRAA AgEDAwIEAwUFBAQAAAF9AQIDAAQRBRIhMUEGE1FhByJxFDKBkaEII0KxwRVS0fAkM2JyggkK FhcYGRolJicoKSo0NTY3ODk6Q0RFRkdISUpTVFVWV1hZWmNkZWZnaGlqc3R1dnd4eXqDhIWG h4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXGx8jJytLT1NXW19jZ2uHi4+Tl 5ufo6erx8vP09fb3+Pn6/8QAHwEAAwEBAQEBAQEBAQAAAAAAAAECAwQFBgcICQoL/8QAtREA AgECBAQDBAcFBAQAAQJ3AAECAxEEBSExBhJBUQdhcRMiMoEIFEKRobHBCSMzUvAVYnLRChYk NOEl8RcYGRomJygpKjU2Nzg5OkNERUZHSElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6goOE hYaHiImKkpOUlZaXmJmaoqOkpaanqKmqsrO0tba3uLm6wsPExcbHyMnK0tPU1dbX2Nna4uPk 5ebn6Onq8vP09fb3+Pn6/9oADAMBAAIRAxEAPwD78/Y9/Y/0fxv4Q1jxP4o1jwv4iuPEOhy6 hDDf3H+maVc5z9onySdmWJLMcEFSQc8eAfFfSvEn7LHjGbwvo3jq7urUxpeNPomoSw2kzuME jY+GYBQCfVcdqTSvitrH7LPxJ8daP4Xkha1uZ7rRJmvIhNJLbJK6bSRgAsAMkAHI4xXd/sif sieG/j58Nr7WNYvtctbq21OSzVbKaJIyixROCQ0bHOZD3xjHFfxHmmNwGKyrCYPL8P7KvHnv PnfR+9qtZcz1974do6H9p5fhMbhMzxeNzHEe1oS5LQ5I9V7uj0jyLT3W+bd6n//Z

--------------GwlPkhZ0dHd5giKQOtFyACWy--

JoMe2704 commented 2 months ago

I would appreciate an "eye-glasses detector".

Actually, this is easy to implement based on the OFIQ code. The face parsing map labels eyeglasses (as 6). You can just check, if a significant part of the eye region (e.g. the EVZMask region defined in EyesVisible) of the face parsing map is labelled as 6.

tadam98s commented 2 months ago

I believe it is an essential fix as the images we are all getting now come from cellphones and not from professional cameras. and cellphones reflect the phone holding position in the jpeg exif flag 274: phone held horizontal left or right, flipped or both. Such is life. So it is best to first "correct" the image to its tru orientation and then do the processing.