[FEAT] immich API

[jsapede]

Is your feature request related to a problem? Please describe.

Thé problem relies on 4 points :

• compreface and deepstack arent updated nor maintained
• compreface can only use cuda Hardware accélération witch is lot energy consumming and limited to expensive hardware
• immich uses various huggingface uptodate models for facial récognition
• immich supports openVINO hardware accélération on low consumption hardware (intel HD)

and finally enven having immich working on facial recognition imposes to have specific container / resources for compreface / codeproject AI.

Describe the solution you'd like
Use immich API for classification and facial recognition
Storage /archiving in dedicated libraries / assets

https://immich.app/docs/api/

[tomerh2001]

Any updates on this topic?

[townsmcp]

I would like to see this as well. But just an FYI @jsapede Compreface can work with Intel GPU for hardware acceleration if you use the right container version

[jsapede]

How do you set it to use Intel openVINO ?
Didnt found anything

[townsmcp]

You have to use a custom build. I find the easiest way is to use custom build with single docker container so it is all in one.
The single docker container info can be found at https://github.com/exadel-inc/CompreFace/blob/master/docs/Installation-options.md#single-docker-container
The custom builds can be found at https://github.com/exadel-inc/CompreFace/blob/master/custom-builds/README.md
To work out the exact docker image is a bit of trial and error by using release info to create the image string. For example, I am using NVIDIA gpu but the image name is exadel/compreface:1.0.0-arcface-r100-gpu

Sorry, does look like Mobilenet isn’t using the iGPU though, however it is using AVX2 cpu instruction set.

[jsapede]

already had a try with arcface 100 but its way too slow if not using nvidia GPU.
mobilenet is way faster but confidence is way under arcface.
however as compreface seems unmaintained and immich is under heavy developement with already interesting features and API i think its stille an idea to add it as recognition engine.
moreover compreface itsel uses lot of harddisk space for limited functionnality

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[townsmcp]

Adding a comment to keep this alive as a request

[PilaScat]

not stale

[townsmcp]

@jsapede i have just asked the question to ChatGPT and got and answer that seems plausible (haven’t implemented it yet). Let me know your thoughts/if you have success?

https://chatgpt.com/share/66f314d9-66b8-800b-b92a-5fc36fc8feeb

[jsapede]

Your exemple Never set how to use to immich api inside, double take

[townsmcp]

Try this:

https://chatgpt.com/share/66f314d9-66b8-800b-b92a-5fc36fc8feeb

[jsapede]

Sée Replay here

immich-app/immich#12275

[townsmcp]

Damn. Sounds more complex, maybe the api will allow uploading into a temp holding area/person but it will all add processing time

[jsapede]

Maybe uploading a file in a specific "frigate" assez and ask api for facial récognition then get back the name from the api

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[PilaScat]

not stale