People and Face recoconicion

[cthomas]

I just moved to IMatch 2020 but I can't get any of the People and Face nonrecognition  to work. And I can't see were the help files shows where to start. Were and how do I start?

[Mario]

I suggest you start here:

https://www.photools.com/help/imatch/#people.htm

It explains how to perform face detection, so IMatch finds faces in your files. You can then see the green face rectangles in the Viewer.
Then you do face recognition which looks at these faces and figures out the person.

I thing I have explained it quite well in that help topic, starting simple.

[cthomas]

I suggest you start here:

https://www.photools.com/help/imatch/#people.htm

It explains how to perform face detection, so IMatch finds faces in your files. You can then see the green face rectangles in the Viewer.
Then you do face recognition which looks at these faces and figures out the person.

I thing I have explained it quite well in that help topic, starting simple.

You did ok  But can you show me some step by step. But can you show  me sometgigh Like a tutorial. I don't even know where to begin?

[ben]

Hey cthomas,
What would you need for a start? Maybe we can help. I think Mario could need some rest. 
Have you read the help?  Read it twice, fist to get an overview, second to understand the details.
Then, select a some pictures with faces, let face detection start and go into the viewer to confirm faces.
Ben

[Tveloso]

I would also add that you should take care to not assign/confirm too many faces from within the viewer, as that creates Trained Faces for the IMatch AI, for that person.  As mentioned in the help, having too many of those, can actually reduce the recognition rate for the person...so after creating a representative set of faces, in the Viewer, additional Confirming/Rejecting should be done in the File Window.

Also, if you did not already have any Face Annotations in IMatch2019, then you will not have any People records in the People View, in IMatch2020...so you can either create them directly in the People Vew (the add person button at the top), or directly in the Person Selector in the Viewer.

The Help of course discusses all of that.

[claudermilk]

It's been a very, very (,very, very...) long time since I've checked in. Life got in the way. Anyway...

I also just moved to IMatch 2020. I read the help a few times, and picked a small group of images to get my feet wet. It's really as simple as picking a few images and fire off the face detection process. If all the faces are not found, I've had good luck with trying again with the next step down for small faces.

I've taken the advice of pre-filling the People section with all those I already have in my Categories. That helps with the confirmation process since they are already there & I can simply start typing in the name.

I've found the AI works amazingly well. It even correctly identified an older picture of my niece at 2 based on my test run of her at 13. =8O  I almost mis-identified her in the older picture. The help is right, after about a half-dozen images the AI is really good at picking the right person.

A side note, I've also found it entertaining with the mistaken identities. Usually family: brothers, sisters, brother/sister, father/son, mother/daughter.

[Mario]

I've found the AI works amazingly well. (...)
A side note, I've also found it entertaining with the mistaken identities. Usually family: brothers, sisters, brother/sister, father/son, mother/daughter.

This is how the AI works. It identifies features like node, eyes, eyebrows, mouth etc. in the face. And close relatives often look "alike" and thus can fool the AI. And, probably, also humans 

[plastikman]

I would also add that you should take care to not assign/confirm too many faces from within the viewer, as that creates Trained Faces for the IMatch AI, for that person.  As mentioned in the help, having too many of those, can actually reduce the recognition rate for the person...so after creating a representative set of faces, in the Viewer, additional Confirming/Rejecting should be done in the File Window.

Can you elaborate on this? I just hop through all my photos in the viewer to confirm every single face in my library. I found the people overview cumbersome because with more people in the photo, it is impossible to tell if the face recognition identified the right person as X. This could be solved by zooming into the face of the ID, like in the new trained faces window.

But I would like the information you state confirmed since I do have like 500 trained faces for someone like my daughter and I noticed that the accuracy of identifying her down was going down recently.

[Mario]

You can easily un-train images in batch via the Person Editor. Click on the smiley icon at the top and then on the ? for usage instructions.
Reviewing Trained Faces and 'Untrain' Faces

[plastikman]

You can easily un-train images in batch via the Person Editor. Click on the smiley icon at the top and then on the ? for usage instructions.
Reviewing Trained Faces and 'Untrain' Faces

Thanks. Was just now reading the updated help section where this behavior is explained in detail. Let me see if I can fine tune. How many images would you recommend per "change" in age. E.g. my daughter is now 2,5 years old. I would say distinct phases are baby, 3-month old, 6-month old, 1 year-old, 2-year old. Do a few e.g., 3 id's for each suffice?

[Mario]

5 to 10 images per person is usually sufficient. A couple more if the age varies a lot (childhood, not so much adulthood).