Setting up a drive to be available via Samba is a relatively simple thing to do. The drawback is that you have files that are as organised as the media asset manager. It can be quite chaotic unless you have someone trained as a media asset manager, archivist, or other, to help order photos videos and more. To some degree Nextcloud is just as disorganised, initially.
I have spent more than five minutes experimenting with Nextcloud through several iterations and I have finally set things up as I want them. I have Nextcloud running on a Raspberry Pi 8GB. I chose this device because it’s the highest spec pi available at the moment without months of waiting. I could have used an HP Elite Book from several years ago but I want a machine that can be on permanently.
The first sync is slow. Twenty hours ago I started with over 19,000 photographs and videos and now I still have 6400 remaining. I activated recognize, an AI solution that recognises music genre, objects, human movement in video, people, bodies of water and more. I also have a tool running that maps photographs to show where they were taken.
The beauty of Nextcloud for photo storage is that it allows you to sync from your phone via the app, but it also allows you to upload photos via a web interface, or if you’re so inclined via file transfers on the back end. I have yet to test the latter. The idea is simple. If you have terabyte drives filled with photos that you have already organised by year, month, day, location, and topic, then that file structure should be recognised by Nextcloud. The work that you have done to organise media assets is already done. It’s just a matter of letting Nextcloud see them, and it will take care of mapping, and recognising objects, monuments, images with people and more.
## Facial Recognition
With time it recognises faces and the faces are just given a number. You can then provide them with a name. I added my name to the collection of photos of me. It needs 120 faces before it starts to recognise individuals. As the model is self-hosted this data stays local to your system
## Object and Landscape Recognition
I noticed that it recognises water, alpine landscapes, signs, boat, bridge, flower, furniture, historic, information and more. The Pi is still working hard to ingest the remaining 5600 photos but when that is done it will have plenty of time to recognise what is in pictures.
When you work as a media asset manager it takes time to tag images, and to add location data. If AI can provide some of this information automatically then it saves a lot of human time. Time that humans can spend adding images to the right folders.
## Folder Structure
As a best practice you should always use folders with year-date-country-event-name-photographer-initials. If Nextcloud is up and running you can rely on Nextcloud but if for some reason Nextcloud crashes, or you can’t use a web browser or app, you want to be able to find things according to year, month date, photographer, and topic. Nextcloud should be an embellishment but good Media Asset Management practices should be prioritised.
To some degree the iOS app can help with this, as long as you set things up properly ahead of ingesting all the photographs. I haven’t seen how to set it up yet, but for now I’m still testing the proof of concept, for mobile phone image backup.
## Using an Intel Machine
What I am doing with the Pi is experimenting with a Google Photo and iCloud replacement. What I plan to do with the linux laptop is use the full power of a normal computer to serve as a media asset manager for when the machine can be turned off, and on, when not in use. The aim of Nextcloud on the laptop will be to provide me with a one terabyte NAS where I can experiment with what Nextcloud really has to offer.
> Tensorflow WASM mode
> WASM mode was activated automatically, because your machine does not support native TensorFlow operation:
> Your server does not support AVX instructions
> Your server does not have an x86 64-bit CPU
When you use the Pi it does not have the required the required x86 64-bit cpu. For that I need to use the Intel machine. It also has GPU acceleration, which I cannot use on the Pi. The Pi is good because it can be on 24 hours a day, as a quick backup tool for your phone, but an Intel NUC machine can be a Nextcloud server with the required hardware to do things much faster.
## And Finally
iCloud and Google Photos are great for backing up when you’re out and about. They’re less great when you want to recover your photos. This is because if you remove photos from iCloud they are removed from everywhere so it’s dangerous to clear photos to make space.
With Google Photos the issue is that their cloud backup solution is 34 CHF more than Infomaniak’s cloud storage solution. It is for this reason that I wanted to have a local backup of my Google Photos and iCloud photos. When I setup the Intel machine and ensure that all my photos are backed up from Google Photos I will be able to purge Google photos and downgrade my Google One plan.
My aim is not to eliminate Google Photos, but to reduce the plan I’m using. I have access to two terabytes but I never use it, and Infomaniak is cheaper, so I prefer to have a single plan. The Intel will be the main backup, and kdrive would be the offsite backup.
In the time it took to write this blog post I went from 6400 images to backup, down to 3800.
Leave a Reply