Requirements and Limitations
Although we put a lot of effort to make frictionless onboarding experience, FaceRecognition is not your average Nextcloud app where you just install it and enjoy it. You will need some beefy hardware, you will have to install some dependencies on your Nextcloud machine, and you will have to be good with command line. With that out of the way, let's start.
FaceRecognition is using deep neural network approach to detect faces. For this reason, machines where this app can be executed should have have really good spec. You need to either have:
- Machine with CUDA-enabled GPU, or
- Machine with at least 2 GB of RAM (4GB is preferred).
Unfortunately, this leaves out Raspberry PIs, lot of VM configurations and lot of home desktop machines. If you are in this boat, we are sorry, but there is no point to proceeding (we do have open issue to support HOG detector on our mind which should reduce machine specs, so subscribe there to be updated).
If you want to leverage GPU, you have to be prepared to compile some dependencies manually (you have been warned:smile:). However, using GPU will mean that processing of one image will take 1-2 seconds, compared to 40-120 seconds without GPU! With lot of pictures, this can mean difference between 10 minutes for all your images and literally days.
Whole FaceFaceognition app is written in a way that you can control when CPU-heavy usage is happening. Without GPU, CPU usage during face recognition also spikes and can hog your machine, so consider utilizing some ways to cap CPU limits, or schedule face detection during idle/non-working hours. If you have GPU, there should be no heavy CPU activities.
There are couple of limitations that you should be aware before installing FaceRecognition application:
- Supported Nextcloud versions are 14+
- Only PHP 7.x is supported (no support for PHP 5.x)
- We provide PDLib packages for Debian/Ubuntu only (if you are not on happy path, you will have to compile it yourself, as explained at Installation page)
- Withouth GPU, face detection can take 2-4GB of RAM and can 40-120 seconds!
- There is no support for shared or external storage, but we are working on it
If you are OK with all this said above, head over to Installation page.
