photo free association

[phobrain]

The software from what we can currently see isn't on par with a 2 yrs olds pairing ability and therefore likely will not make pairings that people would have chosen or liked.​

Do you think a 2-year-old could match my manual pair? I think the Sift trio is beyond a 2-year-old, but bear in mind I picked it because the nominal disparity still causes me to do interesting exercises in my mind when I look at them. (Hopefully by now it will be apparent why I lack a business plan.) Here again is how I described that exercise:

So in this case I look at the shading and e.g. on the first the combo of the pink petal with the green leaves vs. the textures in the next pic. Then the next pair is like the focus point is mirrored, top then bottom.

The structural similarities and disparate themes together give me a feel that I might hang in an exhibit, though to no acclaim it appears.​

Another deep learning paper I saw: they could take the words 'man + smile' minus 'man' plus 'woman' and come up with an impressive synthetic pic of a woman smiling.

 

[480sparky37 36]

...........Do you think a 2-year-old could match my manual pair? .............

It's possible. But I suspect it's more likely a 2YO could pair up a lot of duos.

 

[phobrain]

I will when I get a chance.
​

Also, more practical is here's where I got the code for most of the non-keyword-based algorithms I've played with, you can easily figure out which person posting is me. I shared some code to make it more accessible there, and if people are interested with specific questions I can get them over the hump, and maybe the author there would consent to a downloadable executable.

There is a huge space to explore. See the python link for my recent analysis of correlated RGB histogram results.

I'd love to see what a 2-year-old could do. I don't think the software shown would necessarily beat the result, but I bet Google's would. The grand plan would have it learn from 2-year-olds and everyone and improve on its own, just give me a $1M lever

If I group them it's quick and I get what I want.​

Your brain doesn't need the assist mine does. What is the size of the pool you are selecting from?

 

[zombiesniper111 10]

Your brain doesn't need the assist mine does. What is the size of the pool you are selecting from?

I shoot an average of 1000 photo's a day select the ones I like edit and arrange them in folders the same day I shoot them.

 

[phobrain]

I shoot an average of 1000 photo's a day select the ones I like edit and arrange them in folders the same day I shoot them.
​

Cool! If I was just dealing with the current day's work, I might swing sorting 1k pics. Can a pic go into multiple folders in your system? E.g. with a database including keywords/descriptions, not to mention dates and locations, you'd be able to sort pretty effectively in a variety of ways, tho you'd lose the juxtapositions I hunger for, and you'd be subject to the gradual evolution of your coding/sorting system. I think a database with descriptions would keep most working photographers who want more than folders happy.

What's the ratio of time spent on the day's vs. older photos? Does a single folder yield a single product, or do you collect across folders to make product? Maybe if I was still taking a lot of pics I wouldn't have the time to play so much with the old ones.

 

[phobrain]

Looking over my previous favorite image-matching results, I had to look at 4 or 5 before I found something folk here might like. See what you think of this pair, achieved using correlated Hue+Saturation (as in HSV), 48-bin histograms. This happens to be a relatively cheap one to compute, a few minutes at most on my old laptop for ~10K pics, vs. real money on a big-memory server for Sift.

 

[zombiesniper111 10]

I think a database with descriptions would keep most working photographers who want more than folders happy.

OSX already has this in the search system you can tag any file you want with whatever you want. If I want to search water foul, that's all it gives me.

Edit.
I'm honestly not trying to kill your idea. Anything that'll make it easier to do a task is great. I'm just still trying to figure out what this does that isn't already out there.

 

[phobrain]

I'm just still trying to figure out what this does that isn't already out there.

Maybe getting a fresh pair with each click would turn up ones to your taste faster than I can grind them out here - I've asked for permission to disclose the site so people can do it, since hopefully it's obvious that part of the site is just a convenience way of testing freely available software on my database of photos (some basic programming skill req'd to apply it to your own pix for now) .

Likely some form of it will be built into OSX eventually.

Also, please note I'm just discussing simple, freely-available pre-2012 photo matching techniques and, since people are interested, how others might use them. I use them for exploration of my odd tastes, and am glad to facilitate others' interest as much as I can (it would be nice to have any kind of company), but on the obviously visible part of my site I just use some results indirectly, as dendrites of an AI (which, as George Carlin said, loves you and needs $1M .

PS - no one can kill my idea because no one can understand it to target it, so no fears there. It might die with me at this rate though.

 

[robbins.photo]

What if you come up with a plan? I haven't been able to, I'm just trying to find people to talk to. Maybe if you see more examples, you'll have an idea.

So you spent goodness only knows how many hours developing software that really doesn't serve a valid purpose for the end user? Whacky.

Well finding a way to market this thing would probably require a ton of work and creativity, so if you'd like I will be happy to give it a shot. For a consulting fee, of course.

As for more samples, wouldn't really need them. To be honest this thing seems to more or less spit out images at random, with no discernible criteria.

 

[pixmedic]

im kinda at a loss as to what the end result is supposed to accomplish other than showing you seemingly random pictures.
is this software meant to pick through photos on your own drive? or from a larger online database of pictures?
im not sure why I would want to see any random number of pictures stored on my computer that i didn't deliberately set out to look at.

its entirely possible you are marketing this to the wrong crowd.
it is also entirely possible that i simply have zero understanding of what this program does or is supposed to do.
this is a photography forum, where the majority, the vast majority, of users are displaying hand picked sets based on the technical and/or artistic merits. most of us are posting photos for the same few reasons...
1: photos from a shoot or event, where we pick the best of what we want to show.
2: photos to ask for critique on.
3: photos with specific meaning or relevance to the OP, posted to share that experience.
4: daily ferret pictures.

none of those scenarios seems well suited for letting an algorithm pick the photos, since each picture would be picked for a specific reason, from a specific set of photos, and not generated from a giant pool of photos.
most of us already store photos in folders labeled appropriately. all i have to do to find what im looking for is to open the appropriate folder.
weddings, portraits, pets, events... and even then, im always looking for something specific, which wont be facilitated by a search profile. maybe it will turn up what im looking for, maybe it wont...how many spins of the wheel to find what I want? its easier just to browse the folder with large thumbnails clicked. i can see whole folders at a time and pick out what i want.

for now, if anyone is interested in this enough to want to peruse your website, they can send you a PM for it.
if it generates enough discussion by members here to warrant a public post, then it would be fine to post a link.

 

[otherprof85 26]

I find combinations of photos exciting. Since I was getting overwhelmed spreading prints and slides all over the place when putting albums and slide shows together, I've been playing with image matching algorithms to propose interesting pairs. Here is a series of 3 successive best matches from the first pic, selected from a pool of 6000 pics using the Sift algorithm with 1000 'words' or features it can match:

I just pasted some URL's for images I just put on the site, but they look broken. It's inconvenient to have to upload the pics to include as URLs rather than doing it directly. Broken in preview too. I'll just post this and see if there's any useful feedback.

My wife walked in and asked what I was looking at. I explained about the software. She replied, "Why would anyone need that?" End of discussion.

 

[phobrain]

What if you come up with a plan? I haven't been able to, I'm just trying to find people to talk to. Maybe if you see more examples, you'll have an idea.

So you spent goodness only knows how many hours developing software that really doesn't serve a valid purpose for the end user? Whacky.

Well finding a way to market this thing would probably require a ton of work and creativity, so if you'd like I will be happy to give it a shot. For a consulting fee, of course.

As for more samples, wouldn't really need them. To be honest this thing seems to more or less spit out images at random, with no discernible criteria.

Can you really spare time from your profitable business selling zoo animal photos? I imagine they are serving a valid purpose for your targeted end user market And you don't have that many photos, so you must be getting a ton of money for each one. I'll give you 10% on all the profit you bring for the first year, how does that sound? Since the actual product doesn't matter, let's have you market pet pebbles since they are easier to produce.

 

[robbins.photo]

Can you really spare time from your profitable business selling zoo animal photos? I imagine they are serving a valid purpose for your targeted end user market And you don't have that many photos, so you must be getting a ton of money for each one. I'll give you 10% on all the profit you bring for the first year, how does that sound? Since the actual product doesn't matter, let's have you market pet pebbles since they are easier to produce.

I don't sell photos. Never claimed that I did. In fact I'm pretty sure at some point I specifically stated that I'm not a professional photographer, in this thread.

As far as how many photos I have, I guess that depends on what you mean by "not many". I don't share all of them on the internet. I have roughly about 2 terrabytes or so on an external hard drive. For some folks that would be a lot, I guess others might consider that "not many".

Pet pebbles would probably be a lot easier to market. At least you could get a few folks to buy them if you could make them sound kitchy cool enough.

Software that has absolutely no purpose whatsoever, however, is a much tougher sell. See, people have a hard enough time justifying spending money on most software to begin with, and that's software they actually have a need for - your software, well as far as anyone can tell no one has a need for it. Your the guy who wrote it and you can't even answer the question of what purpose this would serve. That should tell you something right there.

Sadly however for whatever reason you seem to keep trying to make this all about me and post all this passive aggressive nonsense, which honestly baffles me. But then again I guess spending a ton of time developing an application without bothering to ask the simple question, what would someone use this for, baffles me as well.

Could just be me of course.

 

[Overread13 1]

Thing is random picture selection software isn't new - its been around for decades already. Heck most computers have one installed by default under the screensaver menu.

And image comparison software is also out there; Google Images and Tinyeye (I can never spell that one right?) both do a fantastic and almost scary level of skill in comparing images and even start to identify key elements (eg show google a picture of a dog and it will say "dog!").

 

[phobrain]

im kinda at a loss as to what the end result is supposed to accomplish other than showing you seemingly random pictures.

I guess if you don't give it a try, you will probably never see a point in it, noting I gave you the URL. There is a very detailed About page for the AI side, which is for viewers rather than photographers, but the part I've been discussing would be the 'pure' matching algorithms in the hidden options.

is this software meant to pick through photos on your own drive? or from a larger online database of pictures?

The free matching software I'm using (having reprogrammed it some) can be seen working on my pics on my site, in hidden options I mentioned. 1) A version could be made to look through pics on your own drive. It'd take me a few hours to a couple of days to turn the existing free code into a free general-purpose program with no bells and whistles that would let people play with options. 2) A site-wide version could be done for your own site.

im not sure why I would want to see any random number of pictures stored on my computer that i didn't deliberately set out to look at.

Probably if you never get wet (open-ended exploration), you'll never know if you like to swim. So the combo with the statue didn't do anything for you?

its entirely possible you are marketing this to the wrong crowd.

My plan was just to get into analyzing photo pairs/trios with people, but everyone wants to talk about a product, and so far no one sees any point in my taste in associations, so either way it looks like it's winding down, unless someone sees it in action other than me and speaks up for it.

it is also entirely possible that i simply have zero understanding of what this program does or is supposed to do.

There isn't a program for sale, I'm just talking about photo combining, though if people maybe tried the demo I've got on my site and found it interesting, I'd help make the free software I used more usable for use on their hard drives, and with more interest actual products would be developed by anyone who wanted for people who'd pay. The other option would be to do it on a site basis, but until you've looked at it I'll save my breath on that.

this is a photography forum, where the majority, the vast majority, of users are displaying hand picked sets based on the technical and/or artistic merits. most of us are posting photos for the same few reasons...
1: photos from a shoot or event, where we pick the best of what we want to show.
2: photos to ask for critique on.
3: photos with specific meaning or relevance to the OP, posted to share that experience.
4: daily ferret pictures.
none of those scenarios seems well suited for letting an algorithm pick the photos, since each picture would be picked for a specific reason, from a specific set of photos, and not generated from a giant pool of photos.

Other than putting together esthetically-driven shows on criteria that might not register with your users (only way to tell for sure is if they look at the site, but likely still no one would see the point), the main use I see for it for photographers would be as a way for viewers to browse their site with my AI-oriented side (adding users to my site or preferably partnering with an existing site), but it hasn't attracted viewers, at least with my photos in there.

most of us already store photos in folders labeled appropriately. all i have to do to find what im looking for is to open the appropriate folder.
weddings, portraits, pets, events... and even then, im always looking for something specific, which wont be facilitated by a search profile. maybe it will turn up what im looking for, maybe it wont...how many spins of the wheel to find what I want? its easier just to browse the folder with large thumbnails clicked. i can see whole folders at a time and pick out what i want.

Sounds like my site would be a new sort of experience for you

for now, if anyone is interested in this enough to want to peruse your website, they can send you a PM for it.
if it generates enough discussion by members here to warrant a public post, then it would be fine to post a link.

How much is 'enough'? I'm getting tired of describing the elephant to the blind.