From people in categories to Faces annotations

[PierreC]

Hello
I have been using Imatch, for cataloging my private photos, as an amateur, and when I had time.
In my initial data model for catagories, I had created "person on the photo", with as sub-categories the lastnames and as sub-sub the firstnames.
So if I wanted to find the picture for John Deere, I was going into person on the photo>Deere>John Deere

I have now bought Imatch 2020, and I start using Face detection.
What I would like to do is to select the category John Deere, have only those photos; launch a face detection, and then confirm / or correct John Deere as the person on the photos. This would leverqge the work I hve done so far, and accelerate the learning;
Any advise on how to do this at best?

Thanks
Pierre

[Mario]

When you train a few faces per person, IMatch should do the rest?
You may improve your workflow when you work in the File Window in the Category View. If you select the category for "John" and you see only the name "John" in an image, you can quickly confirm it with the corresponding commands in the File Window. Only when there are additional people in the image you need to confirm the faces individually in the Viewer.

[PierreC]

Hello

I have worked on it other the week-end. Please remember that none of my pictures are professional, these are just famiy pictures.

As my pictures are in folders per year, with 1000 to 5000 pictures per folder, I launch the face detection by year-folder.

My feedbacks so far as I did pictures from 2004 (mostly taken with Canon EOS 500), and 2020 from my iphones 6 and 11

For the 2004 ones, the result is good, but still a bit time consumming for me :
-on pictures with more than 3 faces, there is ofter a number of faces not identified as such. So, in the "Viewer" mode, I select "Detect face" and then the proposed "Add new fac annotation", it brings a  new green frame which I position and size above the face missed, and it comes with a "?" in the red square below, I map it to the person. Done
-Then. especially for kid faces, I have a number of cases where iMatch adds the name of one kid already created to a bunch of faces... which ends up with one picture and 5 times the same kid face identified on it.

For the 2020 ones
-after having run the detect faces batch on them (1000 photos), for a number of them, where the face (only one on the photo) looks obvious to my eyes, iMatch does not see a face. But if, in the "Viewer" mode, I click the "Detect face"  in the context menu, it directly finds the face and even the person. I tried from the "View mode" (thumnails), to detect, even pushing it to "Original size", and it does not find it.


Last point : as I plan to have Anywhere running to let the rest of the family safely accessing the catalog, I started to link the detected faces to a category with the name of the person, so that Anywhere could find it even of not yet using the "faces" function. My question Mario is if you intend to implement the "Faces" function in Anywhere, and if so, when you plan it (so that I will avoid adding the categories)

Thank you

[Mario]

Very strange. The Viewer does the same routine you run when you use the face detection in the File Window with the "Optimize for small..." option.
Is this repeatable for your images? If so, please send me a sample via email and include a link to this thread.

[PierreC]

Will do... interested if I manage to make a video of the screen, of what I encounter also?

[Mario]

No video please. I have over 100 unread emails still in my inbox.
If you find an image which allows you to reproduce this, send it to support email address.

[PierreC]

Will look for the image tonight and send it to you.

Did you see my last question : as I plan to have Anywhere running to let the rest of the family safely accessing the catalog, I started to link the detected faces to a category with the name of the person, so that Anywhere could find it even of not yet using the "faces" function. My question Mario is if you intend to implement the "Faces" function in Anywhere, and if so, when you plan it (so that I will avoid adding the categories to each person)

Thanks

[PierreC]

Very strange. The Viewer does the same routine you run when you use the face detection in the File Window with the "Optimize for small..." option.
Is this repeatable for your images? If so, please send me a sample via email and include a link to this thread.

I looked for such behaviour tonight, but did not find any more case. Will capture ont eh fly if see it again