Peter, thank you for such elaborate feedback! Please see my comments below.

The location on the map and the suburb name they are tagged with are quite random.
They are within, say, 20kms. But sometimes it groups 4 photos together which were taken 20kms apart and puts the location as somewhere that almost looks like the weighted average position of the group. In the example below the top group were taken at 2 different sites - the first 4 and the last one in the group were taken in the same place. The rest were taken at another place 2kms away. On the map they show as a group in about the middle. The second group of photos in the screenshot were taken around the same time as the top group (you can see it’s a series of photos of a battery install).

You got it totally correct. The algorithm does clusterization in time and space to assemble photos into meaningful groups that are close enough to each other. And average position of photos in the group becomes an event location. However, like any artificial intelligence, it requires some training in order to come up with acceptable results. So far it was trained and tested on photos of family vacations and other casual photos of the people. Your use case is quite unique and we did not train algorithm for it. It looks to me, that in your case time and distance thresholds should be much tighter.

I will need to think about it some more, but right now it seems to me that proper solution would be some sort of setting in preferences where your could manually adjust thresholds to get desired result. Please stay tuned.

If it matters I’m trying to index 20k files - it is taking a long time! I’ll leave it running for another 24hrs to see what happens.

Yes, initial indexing may take some time, especially if you have lots of videos in your archive. However, this should be done only once, and afterwards should be working like a breeze.

Andrey So far it was trained and tested on photos of family vacations and other casual photos of the people. Your use case is quite unique and we did not train algorithm for it.

My use case is trying to identify photos taken at different sites. For me, these are places that I help out as a volunteer doing maintenance. They are all fairly close (within 5kms) and I often visit more than one in a day. I take photos of things like equipment that is there and later want to go back and see what the model number/etc is so I can get a manual, parts, etc. I was hoping that this app would allow me to look at the photos on a map and then scroll through the photos for that location. It does sort of do that but your grouping algorithm doesn’t work for me.

It did pretty well at our holiday photos but even for them the algorithm grouped places that were too far apart as one event (for my liking).

The idea of being able to select the distance and time thresholds would be perfect. I’d be looking at something like 100m and 8hrs

There may be a market for your software to people who do maintenance or inspections. In the past for my (paid) work I’ve taken lots of photos of various things at various customer sites. They all end up on my phone which gets transferred to my NAS when I get back. The folder ends up with thousands of images all mixed up - personal, work, etc. Being able to view the photos by location would have been super helpful. Being able to then move/copy a group of photos at a location to a new folder would be super super helpful - then I could have moved all the photos for each site/project to the appropriate folder with little effort.

Peter, thank you for detailed response. Adding such option makes lot of sense, I am going to classify this as feature request and will put it into backlog for future versions.

As to moving photos to a new folder - that function already exist. You can select all photos in the event (via context menu or with Ctrl+A) and then move them using context menu.

Peter
I have created experimental version with support for tuning clusterizartion threshold. So far I cannot guarantee it will make it to oficial release or implementation will remain the same. But at least this is the start, and you will be able to test this fucntionality and see if this actually solves the problem.

Since this is experimental version, I will not be publishing link here, I will send it to you in separate email instead. But here how it works. In this version there is new Prefences dialog, and it has “Advanced” tab. Here you can set your custom threshold.

By default it is set to 6 hours, and this also defines the distance threshould using formula maxDistance = X hours * 5 km/per hour (which is walking speed). So far there is no way to set distance separately from time. If you would like to limit distance to, say 100m, then you will need to set time threshold of 0,02 hours (basically 1,2 minutes)

As I said this is very experimental verison and I am looking forward to your feedback, so we can make it better. Please experiment with this version and share what you think.

One more important thing, new setting will only affect clusterization of NEW photos. Existing events will not be changed. So, If you need to re-cluster some subset of your photos, there are few ways to do that:

• Re-index whole resource. Just remove it in “Library locations” dialog and add once again. This will trigger full reindex of whole resource, including rebuilding event clusters
• If you don’t need to reindex whole resource and only looking to reindex small subset of it, you can just rename or move directories you want to reindex. It will have same effect as removing them from library and adding again.