Start a new topic

Editing faces and face recognition

 Face recognition is still bad. I don't have that many pictures yet, so the same few faces show up a lot, and each time, it says there is only one or two people that are recognized, while in fact, each face is included in several pictures. Also, it doesn't match the faces with already recognized faces (at least on mine). Finally, I accidentally typed in a name and hit enter (chalk it up to fat fingers) and now I can't edit that name.


In comparison to Picasa, this thing is pretty bad.


1 person has this problem

Way improved face merge is coming on Monday.


1 person likes this
Is it actually going to recognize faces? By that I mean, it seems to be ok at knowing there's a face by has no idea who it belongs to. All of my pictures generally contain one of four people. Why does it list them all separately and even after merging with the right name it still never recognizes a new picture is person x.

I would have to agree, with Picasa (as an example) it finds so many faces only by naming one. To go through a 1000 pictures with faces with Picasa was done in a jiffy, but with the monument app it would take forever...and ever...and ever. 


Look, sorry to be so blunt, but this was for sure one of the selling points you guys had, and the app is just not even close to as advertised, not by a long shot 

Secondly; You are actually able to input duplicate names, that shouldn't even be possible?!

Yesterday, I managed to load all my pictures on the hard drive that Monument is now organizing. I started yesterday afternoon, Monument is still working on it more than 12 hours later. Also, as a consequence, I have hundreds, perhaps thousands of faces that were recognized, many of which are more than one image of the very same face. Also, now that there are so many faces, trying to do anything with them on my iPhone is so slow as to not be able to tell if it is doing anything. If I were to try to merge faces (now possible with the update), I will still have to pick among hundreds of images PER PERSON (for several people). Needless to say, it is unmanageable. Any suggestions?

Wait till the analysis is complete. If you have, for example, 10's of thousands of pictures, it may take days for analysis. But once it is done, then:


> Edit > Pick several faces (since the analysis is complete, there will be higher numbers under each face) > Merge. Once this is done, follow multiple iterations.



@Monument: Mind writing a tutorial incl pics about this? 


@Toni: We will make a video once feature set implementation is nearing complete. Thanks.

The suggestion "follow multiple iterations" so far implies repeating the same action hundreds, if not thousands, of times. That clearly is not even reasonable to suggest! I see that you are working on a better face recognition algorithm, so can you simply predict when that will be available?

I don't mean to seem to complain all the time and I realize you are working on fixing things and that is good. However, because the way you sold me on the idea in the first place was because I thought all these problems were already solved. I'll stick with it, of course, and am optimistic, even if a bit disappointed that Monument is not yet there.

 

@James: 


I see that you are working on a better face recognition algorithm, so can you simply predict when that will be available?


We are presently testing. So, as soon as a week or two.


The suggestion "follow multiple iterations" so far implies repeating the same action hundreds, if not thousands, of times. 


We took a giant step forward from having to name each one same name to merge, which was a pain. What we have is much better than where we had before. And as we go forward, we are also optimistic that we can make it even easier and more seamless.

With the way to upload photos to Monument, you suggested that one should wait until the analysis is complete. You should really be more specific there and say clearly "do not try to do anything with the monument until analysis is complete. Do not remove it from its energy source, do not attempt to communicate with it using your smart phone. Once the light quits blinking, you may proceed." I put new photos on my Monument three days ago, and as of yet, the light is still blinking. If I try to use Monument on my iPhone, it basically crashes every time. So, until it finishes analyzing the photos, I have to simply be patient. Also, when you said it may take days for analysis to complete, you should say something about the rate of analysis. For example, if you said "It analysis pictures at the rate of about one per minute." then the user can calculate how long it will take to analyze. For me, more than three days continuously analyzing seems like a long time, and I have no idea how many more days it will take!

 

@James:


you suggested that one should wait until the analysis is complete.


We are not suggesting that you should wait because there is a problem. We are suggesting that you should wait to save yourself merging iterations. Think of it this way, if you have 20,000 pictures, you start merging faces at 1,000 into analysis, then you will have a bunch of faces with one or two pictures in them to merge. Then at say 5,000 pictures, you repeat. But if you wait, all faces would be populated with maximum amount of pictures, and there is less merging involved. It is your choice. I'd rather merge chunks of pictures at at 10 or 20 each chunk, than merging chunks of one or two. Wouldn't you?


 For me, more than three days continuously analyzing seems like a long time, and I have no idea how many more days it will take!


We have tested with 20k pictures. It analyzed for days. We have customers who have 200k pictures. It takes longer. But it is very energy efficient, because we are running an ARM CPU (for phones) connected to a wall outlet, and running Linux, not Android. The whole idea with this device is that the photos are analyzed and maintained 24x7, all in the background. All with a few watts of power. Once it is done, it only needs to process new files arriving. And it does not need to be connected to your phone when it is doing it. While you can potentially do this on a full fledged computer far faster, it'd consume a lot more power, and more importantly, if you left it on 24x7, it would not be power efficient.