Jump to content

Does such technology exist that can replace people in upload acceptance checking?


Recommended Posts

Many people have talked about SS using AI to assess submissions for acceptance, but none of the discussions seem very definitive. It is hard to believe such a task could be undertaken successfully through AI automation.

 

I recently uploaded a small number of images and at least 50% were not accepted for various reasons. Some of them included visible logos or trademarks on top of high rise towers (name of the tower), others for brand names in the title or descriptions. One claimed "out of focus" elements, most likely movement. In fact that observation was quite correct, as it was a slow shutter speed on a tripod and the wind was blowing some trees. They were not the subject of the image, just background. Basically travel destination images are probably more sought after in editorials than commercial advertising, but if commercial advertising is the focus, then yes, a lot of post processing is required to remove signs, trademarks, logos, advertising, even on vehicles, including number plates. It would have to be a fantastic image to go to that extreme on the gamble someone will licence it at a rate that justifies the effort.

 

The point is SS have standards to uphold so I accept rejections, even if the reason given is questionable. ROI is time sensitive, so I don't bother with corrections—unless it is just in the metadata. Many end up somewhere else. For instance, rejections due to having no property release on images of buildings. If the buildings are in Australia (my base) then property releases are not required as building design copyright does not extend to photographs of such buildings. SS (operating under US law) may not accept these images, but I can offer them with RF licences elsewhere.

 

The point of this discussion is to demonstrate how complicated acceptance checking is and question whether AI would be up to the task? I would like to put to bed the argument of whether SS does have and use such AI automation?

Link to post
Share on other sites
1 hour ago, Flexigav said:

 I would like to put to bed the argument of whether SS does have and use such AI automation.

From Shutterstock's Q2 2020 results:

Quote

First, we have enhanced our ability to cost effectively ingest content at a lower cost using AI and automation rather than offshore and onshore manual review resources. You can see the benefit of that in our gross margins this quarter.


https://seekingalpha.com/article/4361410-shutterstock-inc-sstk-ceo-stan-pavlovsky-on-q2-2020-results-earnings-call-transcript

Link to post
Share on other sites

SS does use AI, that has been admitted by the new "wonder" CEO.

Quite how it fits in remains unknown.  Clearly its not full AI or full person but somewhere in-between.

The technology is simple.  Plenty of options, they can look for people and if no release flag that up, look for known trademarks and do the same.

They can give an image a noise score (AS does this before you submit to warn you if it thinks an image is noisy or needs a release).  You can rank sharpness easily (I think SS is doing this to an extent hence foliage and other issue rejections).

So on technical standards its actually pretty routine these days.

Its possible in the SS chain, *some* images are fully AI reviewed and a random selection of others are sent to a human with possible issues flagged.  That person can then choose to either exam the image to decide for themselves or just accept the AI recommendations.

Ive had batches of images and videos reviewed in literally seconds - there is absolutely no way those batches could even have been seen for 1 second by a real person in the time it took from submit to approve or reject.  Even a person blindly clicking "OK" wouldn't have done it fast enough.

I've had videos (20-30 second clips, batches of 10-20) all accepted or rejected in 10 seconds after submission.  Not enough time to have even watched a single clip yet alone actually examined it.

Clearly there is some automation going on.

We've had people report some odd rejections that look like AI gone wrong - clouds released for lacking a model release.  Keywords or titles when they're all completely fine and so on.

 

Ultimately though here i think a lot of it is totally random.  You can get 100 images all rejected for noise/focus and resubmitting with zero changes 30 seconds later get them all accepted.

 

Link to post
Share on other sites

I can't help thinking about new technology teething problems that will iron them self out in due time. In the meantime a few babies are being thrown out with the bathwater. I think SS has so many good images in stock, and continuously coming in, that they can afford to reject quite a number that might have been due to AI teething problems. They are encouraging uploads through their new commission framework that rewards more on bulk, while at the same time raising their standard of acceptance using new AI technology. Ultimately they just might maintain a sizable library (or continue to build a bigger one) at higher quality. While this benefits SS, it may lead to a decrease in contributors, however those who continue will be working on getting their volume up for higher commissions.

Link to post
Share on other sites

Having conflict with AI technology certainly can be frustrating. It just doesn't listen to reason. But when it defies logic... now that is bound to drive people away. I have seen images used in advertisements that include 'blur' as an artistic effect to catch the eye (an impression of speed). I'm quite sure any AI acceptance testing would reject such images saying the image is not sharp, or is out of focus... I don't believe AI can distinguish between desired effect and technical issues. Wow factor often defies logic—it is tied in with human psychology that is linked to personal experience. AI would have trouble reading the story portrayed in an image. Perhaps that is why stock libraries deter artistic images, leaving it up to buyers to turn it into art through post processing...    

Link to post
Share on other sites
11 hours ago, Flexigav said:

Having conflict with AI technology certainly can be frustrating. It just doesn't listen to reason. But when it defies logic... now that is bound to drive people away. I have seen images used in advertisements that include 'blur' as an artistic effect to catch the eye (an impression of speed). I'm quite sure any AI acceptance testing would reject such images saying the image is not sharp, or is out of focus... I don't believe AI can distinguish between desired effect and technical issues. Wow factor often defies logic—it is tied in with human psychology that is linked to personal experience. AI would have trouble reading the story portrayed in an image. Perhaps that is why stock libraries deter artistic images, leaving it up to buyers to turn it into art through post processing...    

I hate to be contrary, but blur can be added to a good image, it can't be removed from a bad image. 🙂 Same goes for easy filters and many others. Used to be someone could take a photo and filter it 50 different ways and upload. Even then that was excessive, but got past. Three is pretty much the limit, if you can get them past, here and other places. The whole idea of selling stacks and stacks of blurred images is over. Once someone said "blurred backgrounds sell..." the flood poured in.

Time to find something different that has potential, that everyone else isn't already making because it's a good seller.

Just for some background, 2012 is when SS started with the computer aided reviewing. Maybe before that? But that's when it showed up in the reports after the company went public. (maybe as they went public) But 2012, so this isn't new. Just how much is done by Bots now, has changed. Of course humans cost money, bots just work 24 hours a day with a little bit of tweaking. It Isn't AI, it is Machine Learning. But I know that's not going to change what people call it. So I've personally decided to call them Bots, because that's less smart sounding than the other two?

I'm not sure of the random rejection thing, because I haven't really had problems. Seems I get things accepted that I thought were marginal. I mean small sensor images or screen grabs from 4K. The similar rejections are amusing, because same object in a different setting, if there's some color and shape match enough,  will get a rejection. On the other hand, people who want to move things around, like a table setting or try to make 100 images, by moving the same subjects around "inch by inch" are not going to have good acceptance anymore. Some of the new automated reviews are better for removing uploading in big numbers abuse. Some aren't.

Next time I get a random rejection (how would I know?) I'll just upload over again and hope I get a smarter Bot?

Link to post
Share on other sites
5 hours ago, HodagMedia said:

Next time I get a random rejection (how would I know?) I'll just upload over again and hope I get a smarter Bot?

Hard to tell with a single image but if you get an entire batch rejected with the same reason (or most of a batch) its a sign.  Especially if said batch contains different subjects with different cameras.

The confirmation then is a resubmit without changes and most of them getting accepted immediately.

 

Link to post
Share on other sites
On 2/7/2021 at 3:34 PM, Studio 2 said:

That's one way of looking at it... There are others, such as 'Sstock are discouraging uploads through their new commission framework that pays mainly 10 cents for subs and wide ranging and incomprehensible amounts for O/Ds etc, starting at a few cents, no matter what 'level' you are, while at the same time making the standard of acceptance equally incomprehensible, being based on an incompetent AI technology that is totally random with regard to what it will and won't accept'.

The commission framework with so as you say (10ct subs for almost everybody) is BS. They are dangling the well known carrot in front of everyone's noses knowing perfectly well that the levels won't make any difference and has been stated many times OD's , SOD's and EL's have almost completely disappeared. Not worth the time submitting even once to SS.

Link to post
Share on other sites

Just did an experiment with uploads. I uploaded an image as an editorial. It was rejected: 

"Rejection reasons (1)

Editorial Caption: Caption is not accurately describing the subject matter or is missing required information such as the shoot date, location, or relevant description. Captions must be in English and cannot contain special characters, spelling/grammar errors, or repeat words/phrases in excess."

I corrected this following the Shutterstock example and re uploaded again.

It was rejected again for the very same reason as above. How would an automated bot know if a caption accurately described the subject matter? How would it even know what the subject matter is?

Editorials require a date following a location. Special characters are not accepted in the editorial caption, so I have to assume the date is just entered as digits mmddyy. I may have been wrong to assume that though. Perhaps it accepts certain special characters like hyphens that were needed in the date formatting (mm-dd-yy) causing the auto bot to reject the file for this non-compliance.

These sort of problems take too long to resolve reducing the estimated ROI to unviable. I simply market these media files elsewhere.

Link to post
Share on other sites
  • 3 months later...

More rejects, this time for focus problems. Of course I recheck my work at 400%. If I find no focus problems I assume the AI has faltered. I have two options, resubmit or post to another stock library. If the same image is rejected by more than one library for the same reason I send it to my outtakes archive for future evaluation and maybe deletion. Sometimes the reason for rejection is valid, but not all the time.

Link to post
Share on other sites
18 hours ago, Flexigav said:

I find no focus problems I assume the AI has faltered.

Your photos often include clouds, trees, grass or water- any of which can confuse the  AI review.  As has been described on previous threads consider adding "Selective focus of XXX"  to your description to direct review toward the subject.

Link to post
Share on other sites
1 hour ago, Elliott Cowand Jr said:

Your photos often include clouds, trees, grass or water- any of which can confuse the  AI review.  As has been described on previous threads consider adding "Selective focus of XXX"  to your description to direct review toward the subject.

Yes, I have submitted a 100+ images of regional travel destination vistas recently. Most are outdoor scenery containing all these elements. A number were rejected for motion blur or incorrect focus on the subject. I do not think the AI knew what the subject was, especially on landscapes. Yes I think you are right about water, clouds, fog or mist and unimportant elements in the foreground or background being soft on focus because of limited depth of field. I will submit some trying your suggestion... thanks. 

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...