FWIW Artificial Intelligence Critique

[Bart_van_der_Wolf]

Critiquing an image is a difficult and a potentially time-consuming task. Taste and personal preferences come into play, and not everybody is as methodical as it should be for an objective/constructive critique. The better critiques allow one to at least get some idea whether an image objectively sucks or not, or better yet allows one to improve oneself with the advice given.

FWIW, for this task there is now an on-line AI Critiquing/Coaching engine available:
https://keegan.regaind.io/

And here is a short review of that site.

Cheers,
Bart

[RSL]

I'm ROTFL, Bart. Critiquing an image has absolutely nothing to do with objectivity. It's a purely subjective thing. What I think we're talking about here is critiquing technique. That isn't the same thing.

[GrahamBy]

I did a few tests:

First, my crockery:
"There's just enough blur, I like. The pic is not very exciting… Keep trying. This is at most 4.7/10. Not so good, but I'm sure you can do better! Try again
#boring"

Then some suggestive lips from tumblr:
"Way to capture the light! You really know how to compose a photo. Lovely subject, nicely placed, pleasant detail, and amazing angle. This image is good, but it could be great. This gets a 7.4/10. This is much better already!"

Finally, one of my low-light performance shots of Cynthia:
"The lighting just isn't working. Your composition is a little uninteresting…sorry 'bout it. There are technical mistakes, but it could be worse. Power up your lighting, composition, and colors. This is at most 2.7/10. Now you've got the idea! Feel free to send me as many photos as you want, I'll be glad to comment on them and give you my feedback. After 10 pictures I will evaluate your level.
#tooharsh#messy"

[GrahamBy]

And here is a short review of that site.

Yep, look at the "Hall of Fame" pics at the bottom, to get an idea of Keegan's tastes...

[GrahamBy]

He's not much impressed by reputations, either. This for Helmut Newton:

"What a moment! Too much going on…my head hurts. This pic is just…ok. Don't forget about the background. I give you a 6.5/10.
#messy"

[Rob C]

Clearly, in my amazing opinion, a congenital asshole with a market of idiots to milk.

The entire critique thing is bullshit, second-guessing and, worse yet, a little ego trip at another's expense, possibly both psychologically as well as financially if you have to buy it. The spiritual version of lizard skin camera coverings. Pretentiousness at its worst because it can harm somebody else. How to work something, you can teach; how to see something, you can not.

Rob C

[GrahamBy]

This one at least is funny and free. i think he might have a vocab problem though, he talks about "blur" when it's out of focus areas.

I just threw in a couple of my all time favourite Sieffs, which scored 6 and 6.5, whereas another one of my portraits scored 8.0... within a whisker of Ansel Adams 

So hey, I'm better than Newton and Sieff... (sometimes). I'll take that 

(Here we need one of those icons with me rolling on my back, legs in the air laughing my lungs out  )

[Bart_van_der_Wolf]

This one at least is funny and free. i think he might have a vocab problem though, he talks about "blur" when it's out of focus areas.

I just threw in a couple of my all time favourite Sieffs, which scored 6 and 6.5, whereas another one of my portraits scored 8.0... within a whisker of Ansel Adams 

So hey, I'm better than Newton and Sieff... (sometimes). I'll take that 

(Here we need one of those icons with me rolling on my back, legs in the air laughing my lungs out  )

Hi Graham,

Yes, it's free, and it's funny how an automatic image recognition program can judge an image on 'creative' elements.

The comments with Helmut Newton's image you posted, do seem apt. Cut off feet, busy background, etc., I wouldn't pay top dollar for such a shot and hang it on my wall, just because it has Newton's name on it. I'd keep looking for a better one, or make some myself.

Some of the images I fed to it, got nice critiques, and a few were not well received (and frankly not my best).

I can live with:
"Great color, pretty subject, nicely placed, and pleasant lighting; good. The angle is good and you really know how to compose a photo. You're getting there! 8.1/10 - you deserve it, champ."

  or

"Somehow you managed to save your subject in all that blur. Beautifully placed subject and pleasant lighting; great. You're getting there! Congrats!! You get a 8.3/10."

  or

"I'm allergic to cats but I guess I can comment still. Omg…that detail and controlled, wonderful blur. What a catch, beautifully placed subject, and good job on the lighting. Okay… this is so good. Congrats!! You get a 9.6/10. This wonderful picture earns you the golden ribbon. Congrats! This cat doesn't look so mean, special achievement unlocked!"

  or

"I really like the angle and good use of repetition. This is a solid image. You get a 7.2/10"

  or

"What a catch, and your angle is getting there! This is a solid image. 7.7/10 - you deserve it, champ."

   and how could I object to this:

"Your subject looks right at home and all cozy in your composition. Excellent detail and king of originality! Honestly? This is so good. Congrats!! You get a 9.8/10"

Of course this is just for fun, but it's kind of amazing how it can pick an image apart and comment on the specifics of the composition and recognizes subjects, or even styles.

One of my images got "Is this a Georgie O'Keefe? Strong idea, interesting subject. Lovely subject, nicely placed and great color. Overall, pretty good shot! Without a doubt a 8.1/10".
And when I tried feeding it again, it commented: "Definitively seen that one before, try something new!"

Cheers,
Bart

[GrahamBy]

The comments with Helmut Newton's image you posted, do seem apt. Cut off feet, busy background, etc., I wouldn't pay top dollar for such a shot and hang it on my wall, just because it has Newton's name on it. I'd keep looking for a better one, or make some myself.

I think that's exactly the point: you wouldn't pay top dollar for it. Very many would. It's as subjective as anyone else's opinion.

And just as I would calibrate the judgement of a human by looking at what he or she sees as valuable, I look at what it has ranked in its Hall of Fame. To me, boring crap. But if you want to be guided towards that style, sure, why not?

Oh, looking in the comments on the Petapixel article, someone fed it Baiser à l'Hôtel de Ville. It scored it at 4.6. Ugly background.

The developer responded:

"Yeah Keegan is not always very good with black & white pictures, and may not understand all the subtleties of professional photos.
It's more designed to help "everyone else" take better pictures. If the score is low enough, it tries to provide helpful comments on which axes the user should focus to take a better picture!"

Enough said.

[Eric Myrvaagnes]

C'mon, Bart! Show us the one that got 9.8 out of 10!

I note that Keegan seems to consider himself an expert on "blur."

[Jeremy Roussak]

Well, my pic of Vatnajökull was just awarded 8.8, so it's obviously a sagacious and perceptive algorithm 

Jeremy

[Theodoros Papageorgiou]

I'm very interested in Artificial Intelligence applications, particularly those related with photography. Google's image captioning system, open sourced about a month ago.


It would be interesting to read about the pattern recognition methods that have been used in the development of Keegan. The practical results are currently useless though. Needs much more work, will probably revisit it after a few years.


A few tests with this software:


Test 1 image, Andreas Gursky, Chicago Board of Trade



Result:
"Strong idea, intriguing subject. Nice timing, very creative angle, and beautiful repeating motif! A very pleasing image. You get a 6.9/10. That's a very decent photo for a start! Can you show me something even better?"





Test 2 Image, Henri Cartier-Bresson photograph 'Sunday on the banks of the River'.



Result:
"Nice framing! Yet this pic is just…ok. Let's give it a 6.0/10. Too bad, your first one was better! Maybe another one?"




Test 3 Image, a kitsch image from 123RF.com



Result:
"Pleasant lighting; great. WOW this is intriguing, superb subject, nicely placed, and what a moment! You got it, chap! 8.0/10 - you deserve it, champ. Now you've got the idea! Feel free to send me as many photos as you want, I'll be glad to comment on them and give you my feedback. After 10 pictures I will evaluate your level."

[Bart_van_der_Wolf]

C'mon, Bart! Show us the one that got 9.8 out of 10!

Sure Eric,

Not that special though. Don't know Keegan, maybe he likes animals ...

Cheers,
Bart

[Bart_van_der_Wolf]

I think that's exactly the point: you wouldn't pay top dollar for it. Very many would. It's as subjective as anyone else's opinion.

Subjective? Don't get me started on the marketing of 'Art' ...

Cheers,
Bart

[Torbjörn Tapani]

I don't know what they are doing with the images but if feels like a hoax to get people to upload more images. For what purpose I don't know. The best one I tried was rated 8.7. I was documenting some work in my back yard so I had a few mobile phone shots of a large diameter sewage pipe.

[ErikKaffehr]

Hi,

I have tested on a  bunch of my images and got decently high ratings. I guess the scale is defined by rules of composition and local contrast. A strong coincidence between what I see as my best images and high ratings.

I would say it is interesting…

Best regards
Erik

Critiquing an image is a difficult and a potentially time-consuming task. Taste and personal preferences come into play, and not everybody is as methodical as it should be for an objective/constructive critique. The better critiques allow one to at least get some idea whether an image objectively sucks or not, or better yet allows one to improve oneself with the advice given.

FWIW, for this task there is now an on-line AI Critiquing/Coaching engine available:
https://keegan.regaind.io/

And here is a short review of that site.

Cheers,
Bart

[Eric Myrvaagnes]

After noting the score of the Kitsch photo that Theodoros tested, I really would NOT want to get a score of 8.0/10.   

[Bart_van_der_Wolf]

After noting the score of the Kitsch photo that Theodoros tested, I really would NOT want to get a score of 8.0/10.   

One person's kitsch is another person's 'Art'.

Cheers,
Bart

[Eric Myrvaagnes]

One person's kitsch is another person's 'Art'.

Cheers,
Bart

True.
And it's to Keegan's credit that your owl rated higher than the Kitsch image. 

[ErikKaffehr]

Hi,

It is supposed to be about AI, so I think it cannot really tell kitsch and art apart.

Theodoros test is an interesting one:

• The program is obviously not taking the commercial value of Anders Gurski into account
• We can also see that it is not that impressed by the HCB-image, but again would HCB images rate high on their own?
• The kitch image is awful, but it may have the right subjects, composition and colours
• It's nice to hear that Bart's image got a very good rating.

I have tested with a few old images of cats and got high ratings, and also some wildlife shots and got decently high marks. It recognises cats but likes my grizzly bear and donkey-deer without mentioning the name the animal.

Best regards
Erik

True.
And it's to Keegan's credit that your owl rated higher than the Kitsch image.