Moogle Stiltzkin wrote:another thing i observed, if it flags something as a potential person but is a mistake, you can opt to select it to ignore. but what if you did that by accident? so how do you make this appear again without having to manually find the pic yourself and re-adding it? can it re-index that "ignore" pic again? or is there a checklist you can unignore from? this is one area they probably can look into for improving the facial recog app. because i find grouping pics via deep learning very convenient (once it's up and working), so i would hate it to permanently ignore images without any means of unignoring it for later automation for re-indexing possible facial recog image scans in the future.
This looks like something they more or less built in already. If you are in photostation and you click the three dots button in the top right corner, and then settings, you see a tab called people. Here you have the option to look at the "ignored people" list and there you can choose per picture to "not ignore it", but whether it then indexes it again is the question. It looks to me right now, as if it only runs a picture through the recognition process once instead of continuous, but i'm not sure.
Guapo81 wrote:This looks like something they more or less built in already. If you are in photostation and you click the three dots button in the top right corner, and then settings, you see a tab called people. Here you have the option to look at the "ignored people" list and there you can choose per picture to "not ignore it", but whether it then indexes it again is the question. It looks to me right now, as if it only runs a picture through the recognition process once instead of continuous, but i'm not sure.
you are right. i found it thx
all i found that went indepth into it was this 13:00
NAS [Main Server]QNAP TS-877 w. 4tb [ 3x HGST Deskstar NAS (HDN724040ALE640) & 1x WD RED NAS ] EXT4 Raid5 & 2 x m.2 SATA Samsung 850 Evo raid1 + 16gb ddr4 Crucial + QWA-AC2600 wireless adapter. [Backup] QNAP TS-653A w. 5x 2TB Samsung F3 (HD203WI) EXT4 Raid5 [^] QNAP TS-659 Pro II 1x 4TB HGST Deskstar NAS [^] QNAP TS-509 Pro w. 4x 1TB WD RE3 (WD1002FBYS) EXT4 Raid5 [^] QNAP TS-228 w. 1x 1TB WD RE3 (WD1002FBYS) [^] QNAP TS-128 Mobile NAS TBS-453DX w. 2x Crucial MX500 500gb EXT4 raid1
Asus AC68U Router|100dl/50ul MBPS FTTH Internet | Windows 10, WC PC-Intel i7 920 Ivy bridge desktop (1x 512gb Samsung 850 Pro SSD + 1x 4tb HGST Ultrastar 7K4000)
spec1 wrote:Lastly, I have several users of set up on my NAS. When I log on as a different user the tags do not seem to carry over. This seems to indicate that the tags are not being saved as metadata to the file. I have to go through this very time consuming process of tagging pictures for each user! Does this seem right, am I missing something?
Hi spec1, I am also experiencing the same challenge as you are in terms of people faces tags not shared between multiple QNAP users (for the photos that are also shared between them).
Does anyone know if this specific feature is still in progress or is it performing as designed? If this is as designed it will be a deal breaker to me and sadly I will not be able to use this feature. This is because, as spec1 pointed out, tagging people is a consuming/never ending process and I currently have 4 users on my NAS that do share photos between them .
same as above, it needs to be available for multiple accounts. and other thing I wonder, there is a lot of time in it to fill all the names, where is this db stored to create backups. what if I move to a other QNAP (?)
The program is really worse. It took me 5 hours to tag some tousand pictures with the admin account. I thought other accounts are able to see that, but they are not. Each account has to tag all pictures himself !? Stupid behavior. Does anybody know how I can share my tags with other NAS accounts ?
Is it possible to backup the tags? I lost all them all when I restared my NAS the last time.
Hey guys, seeing this thread is semi-alive, I'm gonna hijack it to ask if anybody has experience with facial recognition and large amounts of pictures? I have around 290K+, and while a couple days back when I installed it everything went smoothly (It took my TS-453A about 3-4 days to scan all the pictures), I was able to see suggestions, tag names of groups of pictures and see all the tagged albums, but after the scanning was complete all my results seem to be there, but inaccessible. When I go to the people tab in Photo Station it loads a bit and then gives me a "Failure" error without any other info, and when I click something else like the suggested tab it just keeps on "loading", and I can't do anything else. I was kind of hoping this would be my holy grail as I've been looking for an program/app which can group my pictures, but it seems unable to handel all my pictures.
Is there anything I'd be able to do to fix this? Or, on another train of thought, are there other apps/programs which can do this on a local storage like a NAS? Thanks in advance!
I have played with Face Detection about 6 months ago but I quit using it. Here is my experience:
I have around 60K pictures with my family and friends and I tagged around 40 persons. In the first instance I tagged about 20K pics and lost about 12 hours but I started having more and more mixed errors tags especially for my kids and their parents ;-p, then I started again from scratch, without my kids and their parents, lost another 6 hours, but the verdict was the same. I observed that after reaching about 500-800 pics per person, the errors rate will increase dramatically especially between kids at different ages and between teen kids and parents. So I tried again, in the second instance I tagged only friends and hadn't tagged any of my family members who actually were in the majority of the pics. The way FaceDetection works is by finding patterns and grouping a dozen of pics by those patterns. You have to assign names to those pattern faces and you will find that in the beginning you don't have too many errors, but as you move on, the error rate increases more and more until a point when it mixes very different faces and then it becomes very time intensive to correct those errors especially when you reach more than 1000 pics / person, because you have to refine again and again a bigger and bigger number of pics.
We are a family with two kids and maybe in 95% of our pics appear at least one of us. And actually now I think that the most useful case would be when you have a visit from a friend or a close relative and you want to show him some pics with him. My recommendation is to use Face Detection for faces that appear in less than 500-800 pics of your total number of pics.
TS-253A (\\SMB file server, CCTV recording, Kodi for Movies, Photos & Music libraries)
TS-112 (RTRR of all shared folders from TS-253A)
Only BIG thing that makes it very frustrating at the moment is the bug already mentioned by moogle, that if you see a mismatch and you change the person of the face match you get a prompt.
It asks you if you only want that single image to change or all the ones linked to the "wrong" person. When you then choose only the single photo it sometimes still links all the photo's to the other person.
So instead of having 8000 recognised photo's of my daughter and 5000 of my son, after trying to alter one photo, I now have 13000 recoginised photo's of my son
And no way of pulling these apart automatically..
Hope they fix this soon though because like I mentioned the feature is actually very nice.
I can confirm this bug. Did you submit it via helpdesk already?
The way FaceDetection works is by finding patterns and grouping a dozen of pics by those patterns. You have to assign names to those pattern faces and you will find that in the beginning you don't have too many errors, but as you move on, the error rate increases more and more until a point when it mixes very different faces and then it becomes very time intensive to correct those errors especially when you reach more than 1000 pics / person, because you have to refine again and again a bigger and bigger number of pics.
this is a good review of the current status.
my expectations for deep learning for facial detection, is that after the initial bump of teaching the qnap false positives and correcting initial assumptions, that the deep learning will become more adapt at making more future accurate tagging (suggestions) which will still need to be verified by the admin to finalize the picture grouping.
Your issue you highlighted aside, my issue with using facial tagging was that when i tagged a bunch of characters, and later noticed a false positive grouped into someones group. i try to relocate that single image. but the problem/bug is it will then sometimes (quite often) mistakenly also transfer a bunch of other images for that group it's being transferred out of even though i specified only move that single image only. This is the single issue that makes the app unsuable for me until this gets fixed
a nice feature addition for the deep learning is to perhaps also include object detection. although i'd be just fine with only facial detection, as long as it's a stable working version which is what we need.
basically this is what happened. rather than come up with a patch to make the existing facial detection workable, they opted instead to work from the ground up and came out with Qumagie to replace photostation face detection.
since Qumagie hasn't yet been released since it's tied to QTS 4.4.x, i haven't had the chance to test it out.
there are some videos though covering what QuMagie is and how it's improved over photostation and facial detection addon.