Start a new topic

The face recognition problem

 Finaly, my Monument seems to have quit crashing (more or less). I had to uninstall and reinstall on my iPhone and iTunes and that seemed to fix it.


BUT, and this is a big BUT, face recognition is still terrible. The main problems have already been mentioned:


1. Faces are too small, often difficult to see.

2. The same face is poorly recognized, so it can come up hundreds of times.

3. Inanimate objects are often recognized as faces

4. The process of merging faces is extremely slow, and that is an understatement.


But, it seems that part of the problem is the lack of information. For example, how long should we wait after getting Monument up and running BEFORE we start trying to edit faces? Is Monument working in the background recognizing faces? If so, should we wait a week, so that Monument has had time to resolve them? If we delete an inanimate object, will Monument learn to not recognize that or similar objects again in the future? If we merge a few faces, does Monument learn how better to recognize those faces?


Also, I recommend that you make more visual feedback. For example, after touching the "Merge" button, nothing happens for quite a while. If the word "Merge" changed color once selected, we would know that it actually worked. As it is, I touch the button, several minutes later, something happens. Until then, the screen is frozen and nothing happens.


Also, why does it take so long? To merge a few faces and then to rename them can take 30 minutes. In Picasa, that same thing was nearly instantaneous.


I recommend that you make a manual that explains these things in detail.


2 people have this problem

how long should we wait after getting Monument up and running BEFORE we start trying to edit faces?


Monument itself boots in around 30 seconds. But checking your disk can take additional time. But once the disk is recognized and connected, you can go ahead.


Is Monument working in the background recognizing faces? 


Absolutely. Your app does not need to be open or connected.


If so, should we wait a week, so that Monument has had time to resolve them?


No need to wait, apart from saving yourself face merging labor. More faces in each bundle, the less work. More faces is a function of waited time.


 If we delete an inanimate object, will Monument learn to not recognize that or similar objects again in the future?


We have not implemented that.


 If we merge a few faces, does Monument learn how better to recognize those faces?


Yes.


Also, I recommend that you make more visual feedback.


Once we complete the feature set, we will start refining the UX and all of the user interface issues.


Also, why does it take so long? To merge a few faces and then to rename them can take 30 minutes. 


That's interesting. It is nearly instantaneous here with 20,000 pictures. It does take a few seconds, but not 30.


I recommend that you make a manual that explains these things in detail.


We have this and this. The second one is the solutions guide, and we have started to grow that.

In fact, if I were to really try to merge faces, I think I would have to spend several days doing only that. The entire face issue is so time-consuming that it really, to me, is not worth the trouble. However, face recognition (and place recognition) was one of the main reasons I decided to try Monument. Now, without a decent face recognition algorithm, Monument is really not worth having. I can do the same things with my computer and an external drive and the old picasa.

 


1 person likes this

We are currently testing a better facial recognition algorithm, and it appears to be a good improvement, but we don't yet have the comparative metrics. Hoping to release in the next coupe of weeks, once we start the refinement cycle.

The recent update has made the facial recognition much worst. my boy's face being recognised as a lady friend's face ... this is really horrible, unbelievable. BOY becomes LADY, totally no logic. 

We have made no changes in face recognition, but we are testing internally a better algorithm. We have introduced better merging and deletion capabilities, which may impact on who is recognized as whom based on your facial collections.

Hi. I have a problem not covered here I think. When I merge a group the poor quality of recognition means I then have to get rid of 60-70% of the photos added to the name. When I go back to ge screen to select faces to name or merge the rejected photos seem to reappear as you'd expect but the reference thumbnail is not one of the photos in the group but rather one that has now been merged with another name. Makes it impossible to work with. The thumbnails need to refresh.

Well Damn!!!


That's a big improvement! HUGE I'd say. 


Now you can preview the faces before you merge. That's going to help with the recognition software moving forward, that's for sure. Not so many false positives. 


Also, it is way less buggy. Still some room to grow on performance, it's a tad slow and still a LITTLE glitchy here and there, but overall, I'd say this is now a totally usable product. 


I just hope that you can soon choose to delete once you complete the Auto Upload, but the Facial Recognition, just WOW, great work. Really impressed. 


One question, there are still a few images that comes up as faces, will you be adding a feature to say not a face or something along those lines or do you just wait until you find a suggested face that is ONLY images that are NOT faces, and delete that one. 


In fact, I just tested that, and I had to report a bug as it crashed when I tried to delete the suggested face. 

Login or Signup to post a comment