Coming from Picasa. Should I have started fresh with facial recognition?

Tony

I’ve been cleaning up facial recognition for about 5 days, close to 80,000 photos. Probably 200 individuals. TonFotos imported the names from Picasa but the number of false positives is never ending. I wonder if I should have let TonFotos facial recognition start fresh, and not let it be influenced by Picasa. The downside is I don’t remember all the names and among the imported names and faces are names I’ve forgotten. I could find the matching photos one by one in Picasa and look at the names. I wonder if I would have been further ahead starting facial recognition fresh. I don’t know how close I am to finishing, or if I ever will finish the clean-up job putting names to faces.

Andrey

Tony The amount of false positives depends on the quality of faces that are confirmed for the person. If it mostly contains obscured, dark or low resolition faces, then program would naturally confuse it with any other low quality faces.

I don’t think it is necessary to start fresh. You still can improve situation by manually marking more faces for that people, but this time on good quality portraits. However, it can easily be the case, when you have what you have, and there is no more high quality portraits for those people. In that case you can just do “Offer less sugestions” by right clicking on the person name in the list. This will stop this flood, and program will only suggest something for that person if face mach is very confident. This change is reversible, you can alsays go back to default mode later on.

Tony

Thanks Andrey. Would it help if among the correctly marked faces for an individual if I marked the poorer quality pictures of them (even though correctly identified) as “ignore this face suggestion” to improve the database of reference photos? Also I have a lot of photos of the sides of people’s faces and not straight on of their faces. Should I include fewer, if any, sideways views?

I’m still thinking of my work flow here. I had been including all photos where a particular person appears, even poorer quality photos, in order to find all photos of an activity or event. As I think about it now, if I just include good quality face photos then if I go to the folder with photos for that event or date, I’ll see the other photos including those with poorer quality face pictures.

Andrey

Tony I can share my opinion, but only you can decide what is best for you. My opinion is that is AI job to help us do what we want, not the other way around. If you want those photos to be discoverable by that person’s name and appear in the personal album of that person, you should mark faces regardles how poor they are for recognition. And AI will have to live with it. In any case, Tonfotos has enough tools to make sure this does not become problem going forward. You can always limit amount of suggestions if you are getting flooded with poor ones. Also, AI is not that stupid, once there will be more good quality references, it will also start offering better faces (if there are any in the library).

LucVA

Hi, same for me. I also used Picasa for a long time. With Tonfotos I feel disappointed, I have the impression that I need to confirm all faces. As an example, I confirmed about 20k pictures of my daughter, and still I get suggestions of a picture of jar of “Nutella”.

Andrey

LucVA Did you confirm those 20k with Tonfotos? Or you already had those 20K?

The thing is, Tonfotos is great when you start from zero, it works like a breeze, you confirm and confirm, since all suggestions are good. However, once you exhausted all good pictures is comes time of worse pictures. Program starts from most similar faces and once no more left is shows less similar, then even less similar.

So the big issue is to understand when to stop suggesting. And this is the problem we did not yet solve completely. For some people it works OK, for some it still contunues to offer when it should stop already, and that is all in one same photo library. I also have couple of people in my library (out of 150+) for whom I still get garbage suggestions, so I just had to use “Suggest less”. For others I don’t have same issue.

So you are right, we are still fighting with this I did not find optimal spot.

Funny enough, we only get complains when users get garbage suggestions. No one complained when program stops suggesting too early and there are still some good faces to confirm. Probably, that is the aswer, I need to think about it.

Tony

Thanks Andrey. I have two people, one with about 6,000 photos recognized and a second with about 2,000 photos recognized. I’m happy to report that for about 12 hours, TonPhotos has settled down with these two people and there have been no more new suggestions or erroneous suggestions. I think I’m over the hurdle. I still have clean-up on a lot of other people but with less photos.

Part of the issue with these two people with 6,000 and 2,000 recognized photos is I have pictures of them from all angles, straight on, side shots, them wearing hats, sunglasses, and for a couple of years, face masks. Where I have straight-on face shots of other people without sunglasses and without masks, TonFotos does a great job and the suggestions are correct. I imagine it’s a challenge when one person has a variety of appearances, trying to come up with criteria to suggest a name . I can instantly recognize them by the clothes they wear, the pattern on their mask, the setting or background so more than the face.

Anyways I’ll continue working through the rest of the exclamation marks. I’ve been marking fuzzy pictures and photos where there is not a good view of both eyes or a clear side profile as “Ignore this face suggestion”. This seems to help with reducing false positives. I think I found the happy medium for my photo set.