pancakes

MicrostockGroup Sponsors


Author Topic: Shutterstock using some kind of software A.I. to review images?  (Read 24519 times)

0 Members and 3 Guests are viewing this topic.

OM

« Reply #50 on: September 23, 2012, 10:41 »
0
I'm sure the "related image" algorithm just uses keywords and if we see something that's really, really unrelated it's most likely caused by the same spam that results in searches pulling up stuff with no relationship to the search term.

At SS I would agree that it's 'keyword' based and I find SS 'suggestions' to be very good generally but at FT I'm not so sure 'suggestions' are keyword based after finding this combination.

I'll say straight off that there are a few keywords in common with the main shot and the eggs on plate with knife vector (cutting, cutting board and a few others) but looking at the images as a whole the 2 images have a few visual things in common: the rectangular background shapes and tones,  the two faces and two egg yolks and the knife in the vector could be recognised as the girl's arm in the main photo.
But whatever algorithms they're using whether keyword based or optical pattern recognition, they're making a total nonsense of 'suggested' in this case. If you're looking for images of a couple in a kitchen, you probably don't want to be presented with most of the 'suggestion' offered here.
« Last Edit: September 23, 2012, 10:53 by OM »


RacePhoto

« Reply #51 on: September 24, 2012, 11:46 »
0
My intention was not to imply any form of conspiracy but merely to highlight the technological inadequacy of whatever algorithm they're using.

And the other option was, they aren't using any of this technology so the whole contention of "technological inadequacy" and the forum debate is just a nice what if scenario?

And for the others who asked: Illuminati, Men in Black, all that secret society stuff that "they" hide from us. That's what AI reviews on SS would be.  ;D

No comment on image searches, go to Google Image and look for similar images, you get some of the strangest things that have similar colors and shapes. And for the examples posted here, I see melons and melons.  ::)

« Reply #52 on: September 25, 2012, 09:48 »
0
Whatever they're doing, it's gotten slower.  I've had some photos sitting in the review queue for 6 days now, waiting for my 'focus' and 'lighting' rejections.

« Reply #53 on: September 26, 2012, 09:29 »
0
Well I got my rejections after a week.  4 of 11 turned down for 'focus' and 'lighting'.

These photos are perfectly in focus, and are as good or better than any I've had approved at SS in the past.  In fact there's no difference between the ones that were accepted and the ones that were rejected.  I hadn't even moved the camera between them.

They were all closeups of fairly small objects. The micro inspectors have always had trouble with macro; SS seems to think infinite DOF is achievable and IS mistakes the actual texture of an object for noise or 'artifacts'.

It does seem likely that they're using software to try and reject on 'focus' and that in these photos the software couldn't find enough of the obvious edges it needs. No doubt it works fine on ordinary subjects, but it isn't sophisticated enough to handle  everything.




« Last Edit: September 26, 2012, 09:49 by stockastic »

« Reply #54 on: September 28, 2012, 15:25 »
0
Cool, I want this software in my camera so I can just wave it around and get pretty shots automatically!  :o

« Reply #55 on: September 29, 2012, 19:46 »
0
I don't think they use software. Shutterstock generally doesn't like images with motion blur. They'll take them, but it's harder to get them accepted.

« Reply #56 on: October 01, 2012, 17:49 »
0
I resubmitted, unmodified, one of the 4 that had been rejected for 'focus'.  In my note to the reviewer, I simply said I'd looked at it closely and believed the focus was good.   

It was approved.

So I'm not crazy, I do know what 'focus' is, and I'm becoming convinced that a first-pass screening for 'focus' is being done - by a piece of software not sophisticated enough to handle all sorts of subjects. 


Poncke

« Reply #57 on: October 02, 2012, 01:49 »
0
Two people, two different sets of eyes. One reviewer didnt like the focus, the other did. Your test doesnt prove anything imo

OM

« Reply #58 on: October 03, 2012, 17:08 »
0
My intention was not to imply any form of conspiracy but merely to highlight the technological inadequacy of whatever algorithm they're using.

And the other option was, they aren't using any of this technology so the whole contention of "technological inadequacy" and the forum debate is just a nice what if scenario?


They have to be using some  technology; after all there's not some geezer/geezeress sitting there behind a machine deciding which 'suggested' to show you when you view an image.

« Reply #59 on: October 03, 2012, 18:48 »
0
They have to be using some  technology; after all there's not some geezer/geezeress sitting there behind a machine deciding which 'suggested' to show you when you view an image.

Huh? Any chance that can be translated into English?

« Reply #60 on: October 03, 2012, 18:59 »
0
So I'm not crazy, I do know what 'focus' is, and I'm becoming convinced that a first-pass screening for 'focus' is being done - by a piece of software not sophisticated enough to handle all sorts of subjects.


It's not software, it's just a very busy reviewer who will be paid just a few cents (6c in 2007 )for reviewing your image and therefore is going to make a very, very quick decision on the technical merits of your work.

It happens to me occasionally and I always know that it is a 'borderline' image when I submit it. It's not that the image is not in focus, the problem is that not enough of it is in focus for it to be a useful stock image (in the rapid opinion of said reveiwer). This is often a problem with macro images of which I shoot a lot. Just shrink the image down to say 3K x 2K pixels and it will almost always be approved.

« Reply #61 on: October 03, 2012, 19:09 »
0
Guys.  This photo is a collection of small objects laid out flat on a light table.  Camera mounted about 2 feet above, leveled with a bubble level.  Small aperture, carefully focused ( in-focus indicator was lit).  Entire subject in one narrow plane at the same distance from the camera.  Nikon ED lens.   At 100%, it's sharp corner-to-corner.  Not one molecule of the subject is out of focus - nor could it be, given these parameters.   

You'd have to be blind.

« Last Edit: October 03, 2012, 19:39 by stockastic »

« Reply #62 on: October 03, 2012, 19:23 »
0
This photo was a collection of small objects laid out on a light table.  Camera mounted about 2 feet above, leveled with a bubble level.  Small aperture, carefully focused ( in-focus indicator was lit).  Entire subject in one narrow plane at the same distance from the camera.  Nikon ED lens.   At 100%, it's sharp corner-to-corner.  Not one molecule of the subject is out of focus - nor could it be, given these parameters.   

You'd have to be blind.

Post a link to the full-size image so we can see. I've had over 5K images approved at SS, with at least a 98% AR, and I consider their reviews pretty consistent. Not 'perfect' obviously as they are using humans and it is a subjective process.

ShadySue

  • There is a crack in everything
« Reply #63 on: October 03, 2012, 19:40 »
0
Computational image aesthetic evaluation is actually a fairly hot area of academic research right now.  There are quite a few valuable applications for this beyond making a microstock reviewer's life easier.  For example, helping an amateur pick out their "best" smartphone photos from a vacation.

A number of high-level image features can be detected and evaluated together to predict aesthetic quality including foreground sharpness, background simplicity, color and luminosity contrast, depth of field, leading lines, symmetry, rule of thirds subject placement, pattern and texture detail, etc.

The current state of the art does not replace a human expert and perhaps never will, but automated aesthetic evaluation is definitely advancing rapidly and should be appearing in consumer applications, search engines and social networks in the very near future. 

Oh, goody. So all our photos will be homogenous. Whoopee.

« Reply #64 on: October 09, 2012, 17:49 »
0
Sorry, had to revive this thread. 

Earlier I ranted about a bunch of recent "focus" rejections which were totally bogus.   I just submitted all of them to DT, and all were accepted.   For me, that settles it.   
« Last Edit: October 09, 2012, 17:55 by stockastic »

Poncke

« Reply #65 on: October 09, 2012, 17:59 »
0
It settled what? Dat DT has different reviewers then SS? I think you are on to something indeed.

« Reply #66 on: October 09, 2012, 18:16 »
+1
I've seen so many tales of the rejected images that get accepted 2nd time around.  Feed the same data into a program and you will always get the same result so highly unlikely that software is making the decisions.


RacePhoto

« Reply #67 on: October 10, 2012, 10:30 »
0
I've seen so many tales of the rejected images that get accepted 2nd time around.  Feed the same data into a program and you will always get the same result so highly unlikely that software is making the decisions.


Well maybe the software has a RNG and it's set to be harder to pass review on weekends? (just a bit of humor)

Yes I agree, it's humans and that's the only way we would get photos from a shoot accepted, rejected and random reasons why, depending on who looks at them. A flawed computer program would be consistently wrong.  :D

I file it under "This is MicroStock" the wild playground of change, bizarre agency behavior and the unexplained. It's the Twilight Zone of photography marketing. It's the Forrest Gump of business practices... You have entered!


Inconsistant is the rule

« Reply #68 on: July 17, 2018, 18:38 »
+2
***** OLD THREAD ALERT!!!!  ***** charliegnomes is on a roll with restarting threads!


 

Related Topics

  Subject / Started by Replies Last post
25 Replies
13331 Views
Last post January 31, 2010, 12:23
by donding
6 Replies
7252 Views
Last post July 18, 2018, 14:17
by NeonRobot
7 Replies
5094 Views
Last post January 28, 2013, 06:53
by indivstock.com
61 Replies
25389 Views
Last post July 20, 2018, 18:54
by rinderart
13 Replies
7277 Views
Last post January 25, 2019, 11:07
by Uncle Pete

Sponsors

Mega Bundle of 5,900+ Professional Lightroom Presets

Microstock Poll Results

Sponsors