An 8-dimensional spiral traversal of image space

I watched Science Friday's "Creating The Never-Ending Bloom" video about John Edmark's spiraling art using the Golden Angle (the design followed by plants that grow around an axis; ~137.5 degrees), and wondered, what if that could be done somehow in image space? I have implemented something, using RGB 2x2x2 histograms (8 dimensions) to represent photos mathematically, taking the angles between photos and the center of the space to apply the Golden Angle, and using the distance to the center of the space as the metric for the spiral. At first, I used the plain 3D RGB angle between the average colors of the photos with the 8D histogram distance, and the results kept turning my stomach, and I thought it was just a fun exercise. Then I tried using the 8D angle with the 8D distance, and it suddenly started making sense, so I think maybe there's something in it.

 

Here's a description of how to access the option and a link to more math, on the Google group where I got the free histogram code.

 

It spirals in or out continuing the direction of the current pair until it can't go forward, then it spirals the other way. The combos feel unlike the ones from the other options (color match, color opposite, random, and keyword match), both disjoint and somehow meaningful in a way I'm not used to. Out of maybe 100 pairs I've looked at, hardly any seem curate-able, but each makes me think. I tried 10 pairs or so just now for something to sell it, but the last is as good as any:

 

Meh! Neither of these photos says anything to me. Not in line, color, composition, or content.

 

I don't understand what you're trying to accomplish with your manipulations.

Of course if you don't like the photos, others could be added..

 

On the pairing, what if you compare it to the random option ('|') for say 20 clicks each? I wonder if a difference would stand out. The pairs alone don't do anything for me that I can figure out, but the series has an odd fascination, since the pairs are all disjoint in a similar way. I'm thinking of making a special page so you could mix and match the angle/distance functions, so people could see if mismatched functions make them as queasy as the one described above made me. I was a philosophy student, if that explains anything. :) Here's the next 137.5 + 137.5 pair in that series.

 

Edited May 4, 2017 by photoriot

I get the impression that something fairly complicated is being attempted here but it's hard to see from the images presented what is actually being accomplished, perhaps because the individual photos are not themselves abstract. The second set has a sort of jangle to it, but it's hard to distinguish sometimes between intentional and unintentional.

 

I think it might help if you could illustrate what you're doing with more abstract originals, or even with drawings at first, or perhaps to go through it in stages.

He's back!

 

Just don't get him started on diptychs or this thread will never end.

Did you watch the Science Friday video? I'm just applying the same math as best I can in photo space, and seeing what patterns emerge. It makes more sense when you watch a few pairs in succession on the screen to see if a feeling of a pattern starts to emerge, which would be in the relative balances of colors. It would be fun to then scramble each image before display so the color balance was the same but the image abstract, and that might help see a pattern. You can imagine the photos as points in one of the sculptures in the video - the experiment is to see if one can sense the structure in it, rather than to generate great pairs, which is of course also nice to have if it happens. For now I'm going to get to know what I have, then see how it works in simple 3D RGB space with averaged colors.

 

The 8D histograms are like you reduce your photo to 8 average colors, and count them like different denominations of coins in a change counter. The pattern of stacks is treated like coordinates x,y,z for 3D, plus a,b,c,d,e to make it 8D. Each photo becomes like a line from the center of the space to its point, and the formula to measure the angle between two lines is simple trigonometry. So I measure all the angles between the pairs of photos, and choose photos at the desired angle that are another step out from or into the center, depending on direction of spiral.

Edited May 5, 2017 by photoriot

Here's one that's suggestive of what a good pair at 137.5 degrees might look like. Given a slight resemblance in shape and bleakness that might characterize a halfway decent keyword or greyscale-based match, the color-angular component is reflected in the distance on the color wheel between the two shades of grey of the photos. Thinking like a program suggesting drapes to offset the rug, based on an article about the Nautilus shell it read in the gym. Since I've looked at 67K color-similar and keyword-similar pairs over the last couple of months (curated pair yield: 15%), this is fun for a change.

 

Here's a classic abstract at last, simple enough so the color wheel thing is blatant, the forms being a matter of luck, and now I'm hooked.

 

 

 

It's the journey more than the snapshots individually though, kind of like this video of 1-second clips from the Pacific Crest Trail.

Edited May 5, 2017 by photoriot

When it gets boring, I ponder how exotic 8 dimensions is, like string theory, yet from our perspective as photographers, it seems totally crude to just divide R, G, and B into 2 halves each. (But going to still higher dimensions has its own problems.)

 

Now I've trifurcated the 'g'(olden) option into '2 3 8', denoting how many dimensions are used for the Golden Angle measurement. '2'D is a perceptually-based space, 'ab' of the average Lab coords of each photo (green-red and blue-yellow). '3'D is normal RGB coordinate space using average RGB coords of each photo, and '8'D is as described above. '2' is the only one that progresses without meandering along extra dimensions, and is the ideal citizen because it uses hardly any disk space, so I will award it both green-violet and blue-purple ribbons.

 

This 2D pair has some subtle interplay that strikes my fancy.

 

I decided to replace the in/out direction of the spiral with my own function based on user behavior, because the directional constraint was giving me too much of the same pics in testing.

 

Here's a 4D pair.

 

I'm still not to sure what this thread is about but can it be expressed in just one photo?

It is about the angle between pairs of photos.

Imagine you are at the center of a circle or sphere-shaped room, and there are two windows 137.5 degrees apart. When you choose an angle option, the room turns in that option's space by 137.5 degrees, and the windows line up with other scenes. The challenge is whether I can develop a feeling for what each space looks like by spinning the room repeatedly, with a little ego gratification because I put myself at the center. I think I'm getting a sense of 8D and 2D, but 3D and 4D are like black boxes - it's like not knowing who your children are. Or a fine artish version of chat roulette.

Where does the 137.5 degrees come from?

Golden angle - Wikipedia

It's related to the Nautilus shell spiral. Also watch the video I link at the top for cool pics of the math in the abstract.

 

I made it so if you click on the top right corner of the pic, it is replaced by one 137.5 further along in 2D, i.e. if you keep clicking there on one pic, it will give you a bunch at 137 degrees, +- 2, so you can compare them to each other as well as the one they are relative to. (Clicking other corners results in color match algorithms being used, and in the center, keywords.)

Just when I thought I was getting jaded, I switched back to 8D and came up with an incredible pair to me, that maybe my greyscale matcher would have put together, but it's unlike anything I've seen with matching of late, and shows that if you beat your eyeballs enough times, something good might come of any method. Since the left half of the pair is by my friends Raf&Skot, I won't post it here, but here's a link. My photo is of arriving in Warsaw from Belarus in the early/mid 90's.

This 8D one I call Wvay grvay.

 

A "dimension" is a property of a single unified continuous entity. But what if you're looking at things that never meet? Parallelisms and/or simultaneous realities?

 

 

 

Cross-parallel slicings can reveal the nature and extent of the gaps between those parallels, but they tell us almost nothing about the simultaneous (discrete) realities themselves.

If you take a set of points in a space and find a center by averaging the points, any line from that center to any point will form an angle with a line to another point, and the calculation of the angle is a known equation. Any form of parallelism beyond that is above my pay grade.

 

4D is gone now, replaced with 27D (RGB 3x3x3) and 64D (4x4x4). Who knows what I'll try next :)

 

Maybe overcooked pasta is a start.

Edited May 10, 2017 by photoriot

Tangled pasta in the garden of 27-dimensional angles.

 

Now that my intellectual lichens have half-digested the quotes, what turns me off initially is that parallelism is merely posited in order to put the writer at the center of his own universe. Also in context it comes across as paralysis through analysis. I am making no claims that these methods I've cooked up capture more than a new, limited snapshot of the 118 million possible pairs of the 15K photos (divided into portrait and landscape sets). If there seems to be inherent disjointness in reality, I take that as a possible indication of our perceptual limitations. I think of each pair of photos as brain probes, and I am trying to map the areas where correlation occurs, in my project of creating a brain from photos. I've looked at about 2500 pairs since I got angular working, with about 500 curated out of that (20%), which isn't bad.

I never found anything interesting with RGB 4x4x4, so I replaced 64D with Hue 8, for an alternative to RGB 2x2x2=8. But it doesn't yield anything interesting.

I trimmed back my initial curations, so in the end it may be closer to the 15% my other options have averaged out to over time. Maybe the rate is dependent on the nature of the photo collection, and none of this matters.. except some options seem to work while others don't.

 

Now I'm amping up dimensionality as far as she'll take it, Captain, giving up on the Golden Angle to explore a shoulder in the angle distribution graph of RGB 32x32x32 (32,768D), around 98 degrees. (Instead of Golden Angle, I'll call it Body Temperature.) The first few results are inspiring, but take minutes to come in as database index building creeps to conclusion. Below is about the 10th pair generated. Watch for the 32K option to appear when it is ready on the site.

 

 

 

That was interesting, but now I switched 32x32x32 to 90 degrees+-1 to see what naked uncorrelation can do.