Google's creatism project


#1

Sometimes the esoteric bubbles up into the mainstream and people go bonkers. Anyway, this might interest a number of people, esp. fans of computer vision.


https://google.github.io/creatism/


(Mica) #2

From a technical stand point this is pretty cool, but when I think about photography as an artform, and not merely as a record, this kind of thing takes the same form as the 1000 monkeys banging on typewriters will eventually produce Shakespeare.


#3

I totally agree. However, in the age of scientism and positivism, the technicalization of everything is inevitable and even welcome. Code monkeys will eventually replace real monkeys :see_no_evil:.


#4

the same form as the 1000 monkeys banging on typewriters

Isn’t that what FOSS is anyway?! :space_invader:


#5

Random keyboard banging
by a FOSS monkey named @afre

We are a thousand little
Monkeys jumping on the bed
With typewriters in our hands
Waxing sonnets day and night
Don’t put us to bed until
We’ve down some coffee and beer
And don’t worry if we fall:
We can always jump back on.

Anyway, I don’t care about being wowed by Google. What interests me about projects like this and from FOSS communities is the thought process that goes into solving a particular problem. As photographers, we are constantly in the business of framing, culling, processing and talking about our photos. We too have a process we’d like to discuss, share and critique.


(Pat David) #6

Link to the paper on Arxiv. It’s a long, detailed read.

I feel like part of what this use-case will likely be as a possible “digtal photo assistant” to maybe help guide new photographers through their editing process in some way (like an AI assistant)? I imagine someone could just throw all the images they have from a collection at the program and it could spit out X number of completely cropped + retouched images?

My only problem with it is the (possible) removal of the learning process by photographers for what actually goes into a nice photograph (classically). It would be awesome as a learning tool, though. Imagine asking the machine about a particular photograph, and getting guided feedback for cropping, postprocessing. It might help accelerate some folks to get a better (standard) eye?

As a learning tool I’d love it. Anything to help folks get into the hobby and accelerate their learning.


#7

Personally, I am always finding fun new ways to cull my photos (not that I take many to begin with). E.g., for community event photos, I sometimes use rudimentary face detection (not recognition) to crop all of the faces so that I can keep track of who is in which photo and whether they look okay or awkward (sometimes those are the best!) in a particular shot. Methods like this don’t detract from the creative process. In fact, they add to it by giving me the opportunity to see things in a different light.

I haven’t found the time to read the paper, which is also linked in Google’s blog post BTW, but when I do, I would be interested in the researchers’ criteria for a well-exposed and acceptable photo and the algorithmic approach they took to select, render and retouch their images.

In sum, it is not only a learning process but having a blast experimenting, remembering the good times that I had at the event and seeing what others do with their images.


(Elle Stone) #8

What do you all think of the actual images on google’s “creatism” link from the first post?

I showed the page to my husband without telling him “who” took the photographs. I just said I’d like him to look through the images and tell me what he thought. Here is a synopsis, pretty much in order of comments as he made them:

  • Great subjects.
  • Good eye for composition.
  • To a certain extent the colors are over the top, a bit overdone.
  • Bit of a one-track mind - any one image, it’s very nice. 2 or 3 in a row, they begin to look alike. Looking at any 20 images, they are boring as h*ll, they are all the same picture.

According to the article @afre linked to (Using Deep Learning to Create Professional-Level Photographs) the “creatism” page showcases the best photographs taken and processed by the google algorithms, rated as “semi-professional” and “professional” by some undisclosed group of professional photographers.

I’m not sure what criteria the “professional photographers” that were consulted might be using to make their judgments. But I think their criteria for “professional” fall somewhat short of the mark of what makes a good image.


#9

@Elle What a fun thing to do. Point #4 was my first thought, then point #3; I tend to have a critical mind :slight_smile:.

I am curious how well the method does under pressure, with varied and difficult samples. The sameness of the source images is boring. I think the professional photographer part is to jump start the tuning process of the parameters of the method, making it semi-autonomous. In any case, Google is always interested in mining user data (insert “the users are the product” trope), so I think they will ultimately be applying that in future products.


(Elle Stone) #10

This article has some nice criteria for evaluating images on both technical and aesthetic merits:

Taking Your Photography To The Next Level (George Barr)
http://www.luminous-landscape.com/essays/next-level.shtml

I think @patdavid points directly to the problem of google algorithms for producing “quality” images (google’s own article uses the word “quality”) by his phrase “better (standard) eye”. There are huge numbers of images on the internet that look just like the google images.

Imagine if google had trained their AI on all the black and white “Ansel Adams wannabee” photographs frome not so long ago. The standard eye would be very different. Imagine if google had trained their AI when pictorialism at its height.

What the actual algorithm can (maybe) teach someone who wants to make better photographs is “how to make more of the same as what google seems to think is good landscape photography”. Now why anyone would want to make more of these images is another question.


#11

Nice read. Feels like “Aim for Excellence”. Many people are content with stage 1; comes with the democratization and wholesale of art and technology and the freedom to express oneself (not so much in some countries unfortunately).


#12

Google’s “cretinism” ??
Not everything understood, because 50% insufficient english, 50% insufficient intrinsic intelligence…
If some people here willing to prosecute in french ->

Computer-assisted conformity and prejudice ???


(Alan Gibson) #13

Eventually algorithms will be built into cameras. I will wave my camera around, it will capture 60 frames a second and choose the “best”, which it will then post-process to make the most attractive image.

Cameras already do some of this. From the rich raw data sources, they make JPEGs with adjusted tones, saturation and sharpening to make the image “better”. Because JPEG is lossy and only 8 bits/channel, these changes are baked-in.

Computational aesthetics is in its infancy. My hope is that it will teach us about how humans measure aesthetics, and open up new possibilities. My fear is that they will narrow criteria and possibilities, like a bad amateur photo club competition judge.


#14

{warning the door of room 67’s open and unattended}

My fear is that they will narrow criteria and possibilities, like a bad amateur photo club competition judge.

That’s the problem with (matching against) a global model/example, with de-context homogenization and old cotton candy.
Be sure we’re gonna have an ever more abundant, retina piercing, soul crushing all-glorious fireworks for the new capitalism decadance era so that everybody and their cousins are able to entertain themselves in stabilizedly capture a 360 8K HDR stream of consciousness flavoured A.I. to choose of the light fields’ essence.
Humans… we… are most incredible and yet so stupid, self-centered, lazy and arrogant beyond wildest imagination. We think our intelect can solve all maladies, fix all limitations… while we starve the chunks of code in us that connects us (emphatic) to the surroundings… we’re not gonna stop till we make the beings that make (all) things for us, then those beings are going to define first what is beautiful then what is good, then they’ll quickly understand that humans are not good neither for vessel earth nor other humans… so those beings we created mirroring ourselves will erase us; hopefully through a beautifully army of merciless chicks, cross breed between Ex Machina, Sean Young, Lang’s the Machine Man / Maria and Her. If Joseph Beuys was alive he would totally jump into a butter swimming pool out of sheer horror.

or maybe not

{closing the door now… squeaky soundzzzs}


#15

Do we really need IA to teach us that ???

open up new possibilities

You can wait for a long time…