QNAP NAS Community Forum

QNAP has enhanced Photo Station with AI-based facial recognition.

As the number of photos on your NAS increases, this can be an exceptionally useful feature for quickly finding important photos of people and special occasions.

An intro to facial recognition machine learning. Pretty broad topic, but as far as use QNAP users are concerned for photostation, using it for photo management unless of course your using the video surveilance apps that use facial recognition for security purposes.
[youtube=]nT_PXjLol_8[/youtube]

[youtube=]Y1lnrGIbweY[/youtube]


I got a ton of images, but i'm a bit lax on file naming them. So facial recognition would help me tag a bunch of images based on the same character for an easy search.

Also helps me find dupes. Though i did use the dupeguru to do this earlier, it doesn't hurt to have a second means for reviewing the images whether there any dupes. for example... i saw some variations of an image that dupeguru did not pick up, because it wasn't 100% the same. so i delete what i felt was unneeded.


few easy setup steps

in app center, install photostation

Go to multimedia management and install the media addon for photostation *to add facial recognition
https://www.qnap.com/en-us/app_releasen ... e=PhotoExt




in multimedia management, you can set realtime scan, or not. also in the media folder tab you can specify which share folder location you want it to be scanning, and also for what media type.



Once you done that, go to photostation. then go to the people tab on the left side panel. in the tag section, if a result for a character shows multiple instances, you can ctrl + left click each of the multiple instances of the same person, then click the "V" then select add tag. This should merge them all into 1 character. Then to select the nicest pic to represent the character, browse the character album for this tagged person, select the image, and select add as image cover for this person.



Some other resources for photo station below.


qnap intro to facial tagging
https://www.facebook.com/notes/qnap/qna ... 311392590/

Photo station overview
[youtube=]-6-Hwhajuys[/youtube]

how to use facial recognition
[youtube=]lE_4fJBoJvQ[/youtube]

how to manage your images on qnap guide
[youtube=]jV6WpBP4i6U[/youtube]

another thing like.... if multiple characters are in a photo situation like this




So earlier i had tagged solution. then from her people album i view the image. then it shows an unknown character, because i had yet to tag it. So i select the facial for it, then from the pull down i select yuri alpha which i had made earlier. Now that character is marked to Yuri. Then a new option appears, do you want to merge the result into the album as well.

Meaning that now when i search under both characters, it will show in both of their facial albums.

I did discover one instance of misidentification. But all i had to do was remove the name tag from the image, and also remove the image from the album. It's still available in the NAS, and you can still later try to tag it. But it will disassociate from the initial album it got mistakenly tagged to.


Also another issue. Notice 2 characters with labels both of which got detected by the facial extension, so was correctly added with the facial box. But then how about the other characters? Clearly it missed them Yes i could manually add in the tags for them, but still the point is the auto tag missed some of the characters in the same picture, probably did not recognize them as faces. So yes they could still improve on their detection rate.

Though maybe the facial extension is more geared to real people, rather than digital art characters. So it would be understandable why the detection rate isn't close to 100% I'm mostly using the QNAP photo station for digital artwork, rather than real life people photos, which it more or less works

And if you weren't satisfied with the original pic for the person in "people" album, you just browse to their character album in people, then select another image, then click "set as cover for this person". Done :}

eventually you will have your image collection properly tagged for easy searching



qsirch also may benefit, making searching images based on their people tag more accurate. machine learning... is this the right phrase ?
https://www.qnap.com/en/how-to/tutorial ... ing-qsirch
https://www.qnap.com/solution/qsirch/en/



Also browsing your collection, if you spot a crappy quality image, you could try replace it with the original sauce
http://saucenao.com

there were also some mistakes where it would suggest create a character label based on an elbow... a leg... torso.... but not the face....


Only solution to that, was to view the image, to know which character it is. Then add the character tag to it.

next step is to go to people tagged, and remove the original box tag.... then re-add the box on the face of the character *which is to left click, drag to form a box over the face, then release. Then fill in the name tag.



Once you done that, you will also have to go to the character tag person album and set it as a picture cover for the character. Otherwise it will show a blank potrait for this character, because earlier we deleted the original tag. it doesn't automatically use the new tag as the replacement, unless you manually re-set it again as the cover.

One suggestion to improve.

Although when i type in the name and it will prompt some suggestions for existing names, I notice that for some partial entries it will not show an existing list of names that may already exist.

For example...

Lupusregina beta was made.

So later i type for a new pic, Lupus but it did not prompt a suggestion maybe i was referring to Lupusregina Beta, to use that instead, and then merge. It did not do that.

Maybe add the name to filter closest to the name you type using a partial.

found another mistake


the same image has 2 facial markers. one of which is a misfire. i just removed the incorrect one.

a few more observations.


trash can/recycle bin in photostation.



as useful as it is, i opted to turn off recycle bin in > options> network recycle bin. Apply. Then clear recycle bin. Then also go to qboost and tick delete recycle bin, clear junk from there. And yet the images in the recycle bin folder was still there.

So then i browsed to photostation, and there is another recycle bin there, and it shows the image content that i thought i had delete earlier (via the steps mentioned above). It was only after i delete via photostation recyclebin in UI did it really get deleted for good.

These 2 recycle bins seem to be for different things? why is that.

In photostation settings you have more options to tinker with.

One such option was an auto save changes to file meta data. I was not comfortable with it, so i left it unticked. Though.... if for example you backup to a second NAS, and later restore from backup, does that mean all that time spent tagging facial is wiped clean because it was not saved?



then there is a backup/restore option. it explicitly tells you it won't backup your people tagging. so then does that mean to save that part, you need to auto save meta data changes in miscellaneous tab?



This portion is where you can see the facial tagging extension status, and also scan location. Though isn't the scan location suppose to be based on the multimedia management location? why is there another location in this section to set? Maybe it's because it uses too much resources, so they let you narrow down which locations specifically to use this feature on. Anyway good thing i'm using a TS-877 so performance isn't a problem for me :} this is one of the main reasons to get a higher end NAS model.

Noticed a new problem

for some reason, a character got mislabeled. So i change the name, apply only for this image. Next it ask me if i want to merge character to an existing one of the same name. So i click yes.

But problem with that is, it would also mass move all the other pictures from e.g. Ainz to the Gutz character (the one i just corrected to) as well.

But if i skipped instead of merge, i get the following which is incorrect.


*note: in a situation where you got 2 persons with the same name, you cannot easily merge them together again.... You have to go about it in a roundabout way e.g. make one of the persons a random name such as "x", then rename back to "Guts", now you get the prompt to merge persons. What would have been far simpler was allow that the selection of both persons in the screenshot, with an option to merge persons, or even a drag drop 1 person ontop of the other. Done!

Maybe to temporarily get around the issue, I avoid using merge for now (to avoid the mass relocation and rename to another character), and instead later merge the similar name tagged together at a later time.


recap:

1. apply to this image only does not always work.
2. merge to existing character has a chance to also merge all other pictures for the person folder where you were renaming the image from, which is unintended. How are you suppose to correct person name and merge, when it will do the same for everything else?
3. even after carefully checking using refresh, the person from another will sometimes merge into another persons collection .....

i'll post a helpdesk ticket later to report the bug.


There is actually a big thread about this here, which other users have encounter some of the bugs i too have experienced
viewtopic.php?f=21&t=116305

the review was done using the latest version for the following apps

QTS 4.3.4.0516 build 20180315 for TS-877

Photo Station 5.7.0
( 2018/05/11 )
[New Features]
- Supports burst photos and time-lapse videos created with Qphoto.
- JPG, PNG, and GIF images can be displayed at full resolution.
- Displays status of instant uploads from Qphoto.

[Enhancements]
- Media folders can be easily hidden.

https://www.qnap.com/en-us/app_releasen ... otoStation

Photo Station Extension 2.2.16
( 2018/05/15 )
[Fixed Issues]
- Continuously displayed "Recognizing the people in the photos…" as a QTS background task on some NAS models.

https://www.qnap.com/en-us/app_releasen ... e=PhotoExt

I doubt facial recognition is really reliable on drawn figures ... maybe super duper deep learning computer farms can base it on some clothing/rest of body recognition .. but whatever simple implementation for facial ID QNAP uses will not do that

dolbyman wrote:I doubt facial recognition is really reliable on drawn figures ... maybe super duper deep learning computer farms can base it on some clothing/rest of body recognition .. but whatever simple implementation for facial ID QNAP uses will not do that

well i wasn't expecting much, because none of their demos ever mention digital art as part of the their facial recognition. but it seemed to kinda work... for example

After testing it a bit more, the machine learning part does sort of work.

Like.... after a while it will give suggestions for possible people the image is suppose to be after i create a facial tag, it will be auto filled by a likely person by the AI. so i guess i need to take more images of the person to improve the accuracy.

But i don't have a chance to do that, because like i said, wasted effort if i can't get persons to stick, if it the bugs ends up moving/renaming to wrong person later. So i've stopped further testing until these issues get resolved, cause no point


Even if auto tagging isn't that good for digital art, can still manually tag facial boxes to link them to person characters. Because all my digital art filenames are a mess e.g.

bad. usually images from imgur albums have crap names by default. unless you bothered coming up with a better one yourself. better. usually direct from artist deviantart or pixiv page. the first example, not very descriptive, so can't quite find images of reoccurring characters except manually. Facial tagging will make it a bit easier. i also found dupeguru to be very useful for cleaning up dupes.

If the auto suggest feature capture rate was even 30-40% to correctly box face for images, that would be good enough for me. I can then do the rest manually.

But i haven't yet seen an example of more than 2 faces having boxes on them. Like the art earlier had 5 or more characters in the picture, but only 2 had boxes. but like you said, if it were real people the result would be much different.

Anyway they explain in 12:17 into the video about how exactly the photo tagging deep learning works. *they're saying the qnap uses the more advanced deep learning method


[youtube=]lE_4fJBoJvQ[/youtube]



slightly off topic, but i wonder if we will get this sort of feature eventually for qnap NAS nvidia card models. we can then do more advance photo editing thx to deep learning

[youtube=]gg0F5JjKmhA[/youtube]

just to confirm, if you go multimedia management and clear your database there, the people tags in photostation "people" will also be wiped.

haven't yet confirmed if misc, save changes to meta data would retain the "people" face tagging or not.

The bug i reported for facial tagging, they discovered the issue and working on a fix. just to update anyone interested to know.