Pages: [1]   Go Down

Author Topic: Searching for surprise - the Excire LR Plugin  (Read 465 times)

peterwgallagher

  • Newbie
  • *
  • Offline Offline
  • Posts: 45
Searching for surprise - the Excire LR Plugin
« on: March 02, 2019, 05:22:13 am »

Experienced photographers often advise us to develop photography projects by looking through our own image catalogs for a hint about what we tend to see, or we like to shoot. Or, possibly, what we don't see or shoot.

It's plausible advice, of course. Knowing yourself is, no doubt, a good way to develop as an artist and casting a cool eye on our own images is one way to know ourselves. Probably.

Still, when I browse the images, most still RAW, in my small catalog (25k images) I don't alway find things that merit a second thought.

I tend to wind through a particular shoot or a series of dates. Sometimes, I need to do this. But when I'm just looking for things that might catch my eye (again), I  find this kind of searching mostly unproductive. It is usually too 'directed' to turn up anything surprising.

I think the reason is, when I'm right-arrowing through the images in a shoot, it's too easy for the remembered context to rob an image of any 'surprise' factor that it might have when seen outside that context. I probably don't see most images with the freshness they deserve.

But when I'm noodling around looking for an idea for a new project I want to look at my photos more objectively: neither skeptically nor defensively. I want to see them with "fresh" eyes.

One tried-and-tested method of finding stuff 'by surprise' is to go to the LR library filters - Text, Attribute, Metadata - and set a few random filters on the whole catalog. Say:
  • The second half of March between 2009 and 2014;
  • All images shot with a lens focal length under 24mm. ]Images shot with a camera I haven't used since... (goodness knows!). All macro images.
  • Images with a square aspect ratio.
  • Images taken in Pakistan, or Paraguay or ...
  • Images with any stars, but flagged as "rejected;
  • Images with color tags whose meaning I've long-since forgotten;
  • Images with keywords "process" or "keeper" or "ugh!"...

Here's another method that I've just discovered.

Download the Excire search plug-in from Excire.com. It's an LR plugin that creates a pattern-matching database of the images in your catalog. Once populated with data about the images, it will identify images that are 'similar' to any given image (and optionally copy the keywords from the selected image to all the 'similars' it finds).

There's a full-function 30-day trial you can install. You have to "initialize" the database by allowing Excire to examine every image in your catalog before the search will work. This can take an hour or so (my case) or several hours or overnight (the Excire recommendation). Once that's done, finding images (10 - 500 as you choose) that are "similar" to any given image is pretty much instant (there's a keyboard shortcut). Excire puts them all in a collection for you.

How "similar" are the "similar" images? Well, here's the thing. Excire gives you no control over that. It seems to determine similarity using color and contrast (?) and content patterns. It's not clear. Nor does the plugin offer any user control over the degree of matching required for 'similarity'.

As a consequence, the matching is, well... surprising. In my experience, photos of objects tend to match well with other photos of similar objects. Excire also allows you to choose faces, including the number of faces and even 'elderly' faces. But most matches seem to be just the sort of thing that a robot might come up with. No surprise there. You can usually sort of see, in retrospect, a possible basis for the implied similarity. But not in every case.

Still, that's a Good Thing. I've seen a lot of predictable pattern matches, especially on shoots where I shot-the-hell out of some subjet to make sure I had a lot of choices when I got back to my desk. But I've also seen many search-result sets that are surprising enough to be 'creative'.

For example, a recent search turned out to be shaped by a color pattern that "kinda, almost" matches in extent, tiling, range and depth a shot of the waters' edge on a rocky beach in southern Australia, an Egyptian tomb-wall decoration featuring vultures, a cactus border in an informal garden, a metal sculpture in the Reina Sophia museum in Madrid and lily-trash floating in a pond in the Melbourne Botanic Gardens.

Now, in that search there were a couple of dozen other images too (several almost identical to the image I started with). But six of the images in this search result - all but two taken with different cameras, spread over nine years including photos from before I got "serious" - were... Well, sort of inspiring as a set.

Inspiring what I'm not sure. I'll have to think about it. As I look at the images, I recognize the similarity. But it's the differences and connections that jump out at me; differences and connections that a search robot knows nothing about (of course) but that are sort of suggestive. Maybe creative. 

What I did was try to reprocess some of the images so that they looked more like they made set but kept their own "personality". Then I saved them as a new collection. Something to think about. Maybe to extend or reconfigure.

I am pretty sure this is NOT what Excire software is really for. I'm sure there are serious professional uses (wedding photos, I expect).

Still, I recommend you try it just for fun.

Excire comes in two versions priced EUR49 and EUR99.

Best,

Peter


Logged

Rhossydd

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2932
    • http://www.paulholman.com
Re: Searching for surprise - the Excire LR Plugin
« Reply #1 on: March 02, 2019, 03:37:27 pm »

Thanks for flagging this plugin up, I'd not heard of it before.

5hrs to index 51k images with a database at 300kb here.
Once indexed the actual searches are very fast.

Yes, I can see how it can trigger some new ways of looking at images. Quite entertaining to play around with to start with.

As ever with this sort of auto identification it gets quite a lot right, but also has it's fair share of howlers*. As such I'd be very wary of using any automatic keywording function because of the risk of it getting it wrong and needing time consuming repairs.

Overall after a very quick play I'm not sure it's something I'll pay for yet. Maybe when I've used it further I'll see more benefit from it, or if it's developed a bit further it's usefulness will increase.
A fun toy to play with next week though.

<later>
*The other concern here is not only does it return images totally unrelated to the search criteria, but what it might miss. Tools like this can be useful if they return too many images, but when they fail to return obvious results that limits your trust in it.

The lack of ability to teach the software seems an omission, but is maybe a big ask.

Another question that I'd like to know the answer to is; 'What is the software adding to image data that allows it to sort images by best match' ?
It seems it might be adding a custom field into the metadata.
I think I would like to have been asked about that addition before letting it loose on the catalogue.
« Last Edit: March 03, 2019, 04:27:04 am by Rhossydd »
Logged

peterwgallagher

  • Newbie
  • *
  • Offline Offline
  • Posts: 45
Re: Searching for surprise - the Excire LR Plugin
« Reply #2 on: March 03, 2019, 05:40:00 am »

Thanks Rhossydd.

Like you, I think of this plugin as 'entertaining', so far.

Most of my images don't have very close matches (unless they're images of the identical object taken, usually, at the same time). So searching for 'similars' is bound to turn up some stuff that is on the margins of 'similarity'.

To me, that's where the fun is. Like everyone, I am used to identifying objects (including from different angles). I can think of e.g. a dozen different domes or spires or trees in my catalog. But I just don't think that much about similarities in colour patterns or arrangements of contrast or angles between the main subject and the subsidiary subjects. I'm hard pressed to remember them too.

My impression is that Excire does a sorta-good job of finding similars based on this sort of thing. The things it finds don't always look that similar at first glance. But I'm assuming (so far) there is some deterministic reason for the selection and that in itself interests me. I can mostly guess what it is. But not always.

Is that interest worth EUR 49 or EUR 99? Maybe not. I'll keep playing for a while before I decide.

On your concerns: I don't think Excire adds anything to the image data. As far as I can tell, it keeps its own database of data about the images. "Best match" is the basis of it's own sort (in the collection) but I cant find see anything inserted into the XMP data for the images themselves.

I haven't considered what it might miss. Obviously a possibility. If you have examples, please share.

Best,

peter


Logged

Rhossydd

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2932
    • http://www.paulholman.com
Re: Searching for surprise - the Excire LR Plugin
« Reply #3 on: March 03, 2019, 06:42:54 am »

If you have examples, please share.
Not sure I can be bothered to wade through 50k of images to find many specifics, but...

One example is searching for 'red' images. I get 112 results (far, far too few for my catalogue). A quick look at one folder of images of tomatoes shows 77 almost identical images of San Mazano tomatoes on a white background, but only 19 of those got picked as 'red'.
Bizarrely about dozen showed up in the search for the colour blue!
Just using a normal LR keyword search found 180 red tomatoes images alone. Add in everything else that's red in the catalogue from Ferraris to post boxes and it's looking pretty poor at finding colours.

Also; Within my catalogue are a few sets of timelapse images. Mostly these are slow changing from frame to frame, sunsets etc. Excire will sometimes pick out some of these images within results, but not all, even when they are very similar. It then also ranks them much more randomly within the results than you would expect.

That it misses out some extremely similar images suggests to me that it might also miss out other images that ought to be considered a good match for the criteria.

Another curiosity is the number of results;
With search by keyword or colour it tells you how many results it has and seems to simply have a finite number of results.
With image based searches you get the number of results you choose to show, regardless of the level of appropriateness.

Mildly amused, but the more I try it the less I'm likely to buy it.
It's really no credible alternative to properly keywording images on import.
 
« Last Edit: March 03, 2019, 07:58:03 am by Rhossydd »
Logged
Pages: [1]   Go Up