Facial recognition thoughts

[cz35895897]

So while browsing catalog discussions (Michel) the subject of json files and facial recognition came up. Yes I also have "tons" of these files (50k+).

I pointed out previously that I had scanned close to 100 years (age) worth of photos and was using the facial recognition to help date faces and events I had only a vague clue as to the actual date. Michel pointed out that he dislikes using the software for such an important task, while at the same time detailing a method to make it better. IE assign a single person to several face ID's. Maybe an age progressive recognition, IE:

• Joe0-1yr
• Joe1-4yr
• Joe4-6yr
• Joe etc etc

Based on all I read this should improve the accuracy and keep the software algorithm from dilution. Of course I would like Adobe to add the Picasa like adjustable thresholds for grouping and matching - wishful thinking I know. But MichelB, if json files are the match criteria, what if we could edit those files, and the software shows us which file goes with what face? Using this method we would be able to determine what a "standard" criteria were for a certain person, and which added photos then "dilute" the criteria because they may be and excessive facial angle/profile/shadow/glasses/etc? Or have the organizer tell us:

"OK I will include this photo under this person, but boy you are really stretching it, the match % is below 10%.Maybe you should assign this photo to this personVariant"

I know this is really deep and a lot to ask for, but it would be similar to what Google Chrome browser includes in their "flags" settings where you can enable experimental settings? And who knows, maybe more people would use facial recognition.

[MichelBParis]

But MichelB, if json files are the match criteria, what if we could edit those files, and the software shows us which file goes with what face?

Sorry, I don't know how face recognition is implemented... The way that information is recorded has changed in recent versions. You had .xml files, now you have .json ones, it seems the faces image data and location on the image are stored in a sqlite database of the catalog like the thumbnail cache. I suspect recent versions are more efficient on big libraries.

Since I don't use it (and don't want to the burden on my otherwise very responsive modest setup) I can only judge by posts on this forum and in the feedback one.

- There are general complaints about slowness especially with big libraries, although it seems that after the initial analysis and indexing things go better, see this today's post:

Elements 13: Max. size of catalog reached? | Photoshop Family Customer Community

- There are many complaints for very high processor requirements, especially for high end configs like yours. Memory leaks? Not clear today.

I don't like to be negative, but I think users expecting too much of face recognition should be warned that they'll have to be patient and better have a fast processor and lots of RAM; they should let the computer work by night... I also understand that a user starting to tag a big library from the start can take advantage of FR. Since my first catalogs (I started with 'Albums', then PSE2...) I have always tagged people manually and very selectively. I suspect most beginners don't realize how fast it is to tag manually after each session.

I just finished editing a session for a big family event (birthdays + new year, 40+ persons). Mostly groups, portraits, duos.. Selection of 210 images among 320. Full tagging keywords, captions. Do you know how fast it is to assign the same 5 people tags to 10 images at the same time? If you do, that makes a huge time difference. As usual I did use wrong tags for some persons... but finding and correcting was fast.

I am not a fan of geotagging, but that's amusing... I still keep my old 'standard' keywords hierarchy (my custom sub-hierarchies) as well as the geotagging feature.

I wonder if the answer to your question would not be to keep both people systems together in the same way. The 'special' FR tags for fast tagging of a big library. Based on that automatic tagging, using a more advanced persons identification: first name, name, married name, nickname, family branch or sub-branch on the standard 'media' tags in your own hierarchy. And only for people you care about, not for adding a host of new Facebook friends...