Shared ML-data for external libraries in multi-user setups
I have checked existing GitHub issues and did not find this addressed.
I propose a shared ML-data feature for external libraries in multi-user setups.
Scenario:
- Family setup: 8 users, 1 external library (~85,000 photos) on Synology DS923+.
- ML scan (faces/objects) takes ~1 week on 3 CPU threads per user.
- Currently, scan must be repeated per user, making it impractical.
11 Replies
:wave: Hey @Tage Tunenet,
Thanks for reaching out to us. Please carefully read this message and follow the recommended actions. This will help us be more effective in our support effort and leave more time for building Immich :immich:.
References
- Container Logs:
docker compose logs docs
- Container Status: docker ps -a docs
- Reverse Proxy: https://immich.app/docs/administration/reverse-proxy
- Code Formatting https://support.discord.com/hc/en-us/articles/210298617-Markdown-Text-101-Chat-Formatting-Bold-Italic-Underline#h_01GY0DAKGXDEHE263BCAYEGFJA
Checklist
I have...
1. :ballot_box_with_check: verified I'm on the latest release(note that mobile app releases may take some time).
2. :ballot_box_with_check: read applicable release notes.
3. :ballot_box_with_check: reviewed the FAQs for known issues.
4. :ballot_box_with_check: reviewed Github for known issues.
5. :ballot_box_with_check: tried accessing Immich via local ip (without a custom reverse proxy).
6. :ballot_box_with_check: uploaded the relevant information (see below).
7. :blue_square: tried an incognito window, disabled extensions, cleared mobile app cache, logged out and back in, different browsers, etc. as applicable
(an item can be marked as "complete" by reacting with the appropriate number)
Information
In order to be able to effectively help you, we need you to provide clear information to show what the problem is. The exact details needed vary per case, but here is a list of things to consider:
- Your docker-compose.yml and .env files.
- Logs from all the containers and their status (see above).
- All the troubleshooting steps you've tried so far.
- Any recent changes you've made to Immich or your system.
- Details about your system (both software/OS and hardware).
- Details about your storage (filesystems, type of disks, output of commands like fdisk -l and df -h).
- The version of the Immich server, mobile app, and other relevant pieces.
- Any other information that you think might be relevant.
Please paste files and logs with proper code formatting, and especially avoid blurry screenshots.
Without the right information we can't work out what the problem is. Help us help you ;)
If this ticket can be closed you can use the /close command, and re-open it later if needed.
Successfully submitted, a tag has been added to inform contributors. :white_check_mark:This has been discussed at length before and is not so straightforward
Request:
- Store ML results centrally per library, so all users share the same recognition data.
- This would drastically reduce CPU usage and make Immich more practical for families.
Thanks!
I am relative new to Immich.
It is very easy to setup, and the ML and AI as such is formidably good.
But the honeymoon is over, and the other users should be let in now, so to speak. Lights are just lit on the christmas tree.
Then - we learnt about this caveat or whatever we should call it:
In my setup we are 8 users in the family. So, if we should redo all the ML, it would consume time, and be a waste of energy and computer power. Alternatively, we will have to stick to one admin user shared by all. No security. So after having done all that effort, for now it is very disencouraging to meet.
Do you therewith mean, that it will not be considered being put on the roadmap.
It won't be on the roadmap, but that doesn't rule out that it might happen some time in the future
Not anytime soon though
Very sad to learn that. I had set it all up, and had created the first non-admin user, and when she logged in, she found that she had none of the common image-files and nothing whatsoever. Then - via chatGPT - I learnt about this policy. It is in stark contrast to the eloquent "welcome story" about a young new family with their first baby that decides to make a family photo album. Very touching, indeed.
The features and machine learning and face recognition, are second to none. But the system as a whole, is useless, if I can't share it with anyone in my nearest social sphere, my family.
The reason for this I just don't understand. Because, now I only see two alternatives:
a) Let all 8 members of the family, which I all trust, share the same admin user account. Good-bye security. Just one of their devices need to be compromised in order for it all to break,
or
b) find another solution out there
We must have a common photo album in the family, if we don't accept to see the cutural heritage, in our very own lives, here documented in our common photo albums, vanish in the digital disruption. It is really important, and disturbing to learn that this is not prioritised
I think it is fairly safe to say that there will be sharing improvements eventually. It's far from a simple feature though and currently our efforts are primarily on a stable release and more fundamental improvements
To reach any of those sharing improvements we need to rewrite significant parts of the server, and rethink our current asset ownership model. Especially around faces there are many edge cases
To be fair though, when I first started out on this journey, I didn’t consider having facial recognition at all 😃 sorry that you don’t find the current state of the application helpful. Our priority right now is to hit stable release before going back and implementing more enhancements such as sharing in this case
A stable platform must always be goal no 1. Of course.
I have tried face recognition in Synology, and in Photoprism. Both are way below the precision I have seen here.
I would not trust to let the tech giants, be it Apple or Google, have the many hundred faces, this system has identified. And - at least Googles - ML on scenery identification is very poor and limited. Here I found, that I can select "dog", better still "dog and beach", or "dog and snow and evening" - and the selections adjust accordingly - really beautiful. Or "funny faces", or "sad faces" or "deep thinking faces" - with brilliant results you can be very proud of. For comparison I tried the same on Google (without having named the faces to Google) - and it simply can't meet that, far from. So 4 or 5 stars to Immich for that being achievable. Only the gallery resulting from such a search needs to be sorted, by shooting date/time or so.
I set out with Immich proceeding from a list:
* Self-hosting on NAS equipped with Docker
* Security incl. reverse-proxy
* Concurrent Multiuser shareability
* Facial recognition
* Scenery recognition
* showing on map and possible geotagging for non-tagged images
* Uploading automatically from multiple mobile devices
* Danish language (I can contribute in that, too)
I have had Synology Photo for a while, but in fact it does not practise multiuser as I wished it, and its AI is too poor.
So, I looked and found
- Immich
- PhotoPrism
- NextCloud Memories
I started setting up a docker with immich, and then left it for 2 weeks, and then tried PhotoPrism. But it is too old-styled and especially its AI (scenery detection and face recognition is considerably under Immich's). I sniffed a bit to NextCloud Memories, and - TBH - I shall do so again.
But Immich's freshness, the engagement all that there, is so impressive and encouraging. And the AI is second to none compared to the others herementioned.
However, concurrent multiuser ability is a must to me. We are a very active family and share a lot of things, including music and pictures.
I was naïve and expected Immich to have that. Also this expectation was strengthened by reading the touching welcome story about settling a family.
Sharing w/o each individual user I share with must do a 2-3 week Machine Learning all over again, with 80% constant CPU load on the NAS server, is really important to me - but I would expect it to be so for hundreds of others, too.
As it is now, I can only share the images in individual albums, not the highly rightfully appreciated AI wisdom tied to the images.
@Tage Tunenet I agree that this is important.
Overall, more flexible sharing and family sharing is a much-requested feature. See for example this FR having 100 comments - https://github.com/immich-app/immich/issues/12614
And no, leaving another comment on it without adding anything new to the conversation (which I believe is hard at this point), doesn't make it come any quicker :P