Face Recognition on dogs

[Rene Toepfer]

Hi,

since today I am evaluating IM and one of the key features is the Face Recognition (FR) for me. As far as I FR understand it creates a hash of the face and compares faces with them. But does such an algorithm also consider human physiognomy? If not then it could be possible to compare also animals like dogs. Does anybody tried or uses FR for dogs within IM? If so how is the experience on that?

René

[BanjoTom]

I tried, since I have a few hundred pix of various family dogs.  Did NOT seem to work...  Even with a couple good full-frame shots used for TRAINED faces, others didn't seem to be detected... 

[Rene Toepfer]

Thanks for your response and explanation of your experience. I work with breeders (and we own also dogs) and therefore it would be nice that in future versions of FR also animals, like dogs, will be detected. My Sony camera has this on a rough base for eyes. So it seems that it is not so easy than I thought.

[sinus]

I think,
humans and dogs are very different for face recognition.
I mean, then we have other animals, what would be closer to human, like monkeys.

Maybe there is some software out, what can do this, but I doubt it. Of course, it is a question of time.
In some years this will be also possible very good.

[Rene Toepfer]

I think so too. The statistics how a head looks like is completely different between human and dogs.
My questions was just for information and with the responses I do not need to put effort on it. It is not an issue which is a disadvantage for Imatch in my use case.

[Mario]

The IMatch AI has not been trained to detect faces of dogs, cats or animals.
It has been trained to detect human facial features and to recognize persons.

The available training data for animals is minimal and often under restrictive licenses (e.g. usage only for educational or research purposes).
While the big players (Facebook, Google, ...) have spent hundreds of millions on face recognition technology to detect humans because they can monetize it (sell more ads, learn who you know and who the people you know know...), there is far less interest in dog, cat or animal recognition.

I'm sure there is research done to classify animal species in photography. Maybe even sub-species detection, to tell a Schäferhund from a Dackel.
But actual face recognition for individual pets seems more like a niche market to me...?!

A network to detect cats/dots/horses or other animals would have to be be trained separately. A cat face looks totally different than a dog or bird or horse face. Very different symmetries, eye arrangement, face shapes. This is high-end research.
You cannot really create a network that can detect dots, cats or other pets and people at the same time and equally well. Results would surely be far less than satisfying.
Probably a software combining multiple separately trained networks, applying each to each image and than determining the best matches. Runtime would be bad, and results would vary a lot. A pre-classification by a human ("Search for dogs in this image") could be done, of course. This could avoid cccidentally tagging aunt Paula as 'Boomer'.

While this is surely a niche market, you still may be able find a commercial service which specializes on this.
Have you tried to the free test sites offered by Google or Microsoft to check if they handle animals? Or searched for companies specialized in this? There are many cat/dog/horse breeders, maybe there is a market for this.
If your camera recognizes your pets, does it store the data in the face region in XMP? IMatch would use that automatically and create "persons" from it.

I'm sure you understand that I don't have the time to invest in research in such niche areas.
I'm just one person with limited time and resources. My time is best spent at enhancing IMatch for the majority of the user base. Things many people need or features which allow many people to do more with their images.

Adding human face recognition on the level IMatch offers did cost me one year or work and a lot of wasted weekends. And many new gray hairs.
The results are astonishing and face recognition and the advanced features built on that (which will be further extended for IMatch 2021) are a very welcome addition - for most of the IMatch user base.
Recognizing specific pets, not so much.

[jch2103]

I believe I recall reading that someone (a university?) was doing research on face recognition for dairy cattle (to track them for individualized feed recipes. milk production, etc.). Sounded like it was in the very early stages, but there was a commercial benefit involved.

[PandDLong]

This is a good opportunity to commend Mario for an excellent job on the facial recognition capabilities in iMatch.

I just went through an evaluation of several all-in-one photo tools and DAM products and many of them had facial recognition.  This is a feature that I think will be very useful to me so it was part of my search and evaluation.

I created a test set of photos and iMatch came up clearly as #1 in this capability.  The accuracy was excellent - even across some wide gaps in age for a couple persons and a pretty limited set of photos to use as the base.

I was impressed - especially given that Mario is but one person.

[Mario]

Thanks. Very much appreciated   

My research of course was based on the research of others. And all this is still ongoing.
Many of these are FOSS projects, but I donate some money to them every year (as I do for ExifTool and other FOSS projects).

My focus was to do something that works well enough for the majority of IMatch users, works locally (no cloud, no privacy issues), with a tight integration into the IMatch ecosystem plus additional useful features - from filtering to searching to sorting to variables. Support for standard XMP metadata of course, so users can use the face data also in other applications. No lock-in!
IMatch 2021 will even extend this even further

I have to say that I'm still a bit proud of how it all worked out.
Any I still remember the evening when it worked properly for the first time (in June 2019!).

[jch2103]

+1!

[BanjoTom]

+1 from me, too, for Mario's work on facial recognition.  I really don't NEED it to work on dogs or anything else!! 

[Tveloso]

Yes indeed +1

Mario has said that no Face Recognition can be perfect...but I don't know - he's made it so that in IMatch, it darn near is!