My research task is face recognition in cars with using deep learning method. Actually, in example we set an driver randomly and then the question is: Is this person driver or not?

So i created an custom dataset that includes 47 different persons's faces when they are driving cars. Every person's have 3000 face images and I used MTCNN for detect faces from images.

But I'm confused; is the problem multi-class classification or binary? I mean the result is binary (driver or not) but we'll set driver after created this model. for that I can not divide the dataset like driver and others.

When I researched I found that one-class classification approach. Deep One-Class Classification Is this help me or not, I don't get it.

What I should do?

  • $\begingroup$ "I mean the result is binary (driver or not) but we'll set driver after created this model". Can you explain a bit more? Is your dataset all examples of "driver" with no examples of "not driver"? $\endgroup$
    – zachdj
    Mar 15 at 16:55
  • $\begingroup$ Actually yes, all persons in the dataset can be set as driver. For example let say person_5 is driver so all the other persons are could not driver. When we gave an image except from person_5, it must say "not driver". $\endgroup$
    – mmevy
    Mar 15 at 17:15
  • 1
    $\begingroup$ Oh I see now! I believe this problem would best be described as "image similarity detection". Given a reference image (e.g. person_5), the model needs to determine if future images match or not $\endgroup$
    – zachdj
    Mar 15 at 17:48
  • 1
    $\begingroup$ @zachdj Oh it related to Siamese Neural Network I guess. I'll be looking at. Thank you! $\endgroup$
    – mmevy
    Mar 15 at 19:25
  • 1
    $\begingroup$ check out this work on facial recognition github.com/wesleylp/CPE775 $\endgroup$ Mar 17 at 3:39