[dsmithhfx]

Creatism

[victormeldrew]

Seems to have a bit of a fetish for mountains.

[PJ1]

An artist (painter) can put together an image out of his/her head. It consists of remembered bits and pieces from past observations. With a bit of imagination, a good artist can assemble that into a meaningful and appealing end product. This computer software does the same thing drawing on stored image segments (make the mountain bigger, steeper, and more snow; put a cow here, etc). But the software is certainly impressive. It goes beyond substituting a sky or adding in an extra cow, which can already be done in post processing if that is your thing.

Seems to have a bit of a fetish for mountains

A very astute observation. Some years ago there was "art" being flogged door-to-door in Australia that was mass produced in Southeast Asia. It still turns up at market stalls. These "original oil paintings" were competently executed, but what they lacked in soul they made up for in mountains (and lakes). People seem to like mountains (and also lakes) as subject matter for holiday snaps and for their walls. Maybe that is because these subjects provide big focal points. For an artist creating a picture, they are broad brush components which are relatively easy to detail. It may be the same for this software. But if that is what many people look for in a holiday snap does not mean that they are crying out for software that can turn out ersatz holiday snaps. This software will have its uses but I can't see it posing any threat to photography.

[stevebrot]

An artist (painter) can put together an image out of his/her head.

Composing from the mind's eye...trees, grass, and rocks plus lakes and rivers and a snug warmly-lit cabin ==> instant Thomas Kincaide. He turned out thousands of very salable images using a fairly simple formula.

These "original oil paintings" were competently executed, but what they lacked in soul they made up for in mountains (and lakes).

In this part of the world they are usually not from China, but are created instead by so-called "starving artists"...people who have skills with paint and canvas and who might turn out five or six identical paintings at a time using multiple easels. Lots of mountains and lakes and sometimes seascapes.



Steve

[35mmfilmfan]

People seem to like mountains (and also lakes) as subject matter for holiday snaps and for their walls. Maybe that is because these subjects provide big focal points

I always thought it was to distract them from their hum-drum, boring, day-to-day lives and surroundings. Isn't that why people travel ?

[PJ1]

Isn't that why people travel ?

That too. What I was referring to was the kind of snaps they like to have to remind them that they escapes from their boring, humdrum lives. But I suspect you knew that!

[pentageek]

Seems to have a bit of a fetish for mountains.

Not really - authors of the bot fed it with panoramic shots from Google Street View, so it is their selection, not bot's. The bot was tasked with cropping, saturation, and dodging/burning, and the fact that pictures look pretty good (i.e they are not total mess) is impressive. Of course for every photo taken and used as input, a human had to decide that given subject was worthwhile. The point of whole exercise as I see it, was to show that 'quasi-intelligent' manipulation of pictures by an algorithm is possible. Last, but not least, machine learning process was possible because human photographers evaluated pictures created by the bot.

[BrianR]

I've just skimmed the article so far, but it looks like they fed their machine a bunch of popular landscape images from 500px, with the goal of "colorful professional landscape(s)". The threw out images below a minimum colour saturation level. As pentageek mentions, the shots are from Google Street View, and their source panoramas were selected mostly from foot paths through mountains. So yes, the types of images were biased from the start. Feed it a bunch of classic B&W landscapes and you'd get different results.

It's pretty neat, but I kinda wish they had taken it furthur. Team up with researchers working on artificial social media interactions (surely someone at Google is doing this) and make a (mostly) artificial photographer profile on 500px. How popular could a Google bot become on 500px? Pair it up with a Google smart car equipped with an AI drone so it could go out and take some photos of it's own instead of cropping out of panoramas. Maybe task it to join a few forums to argue about what brand of cameras are the best. It may end up a sad commentary on how many humans use social photo-sharing websites, but would have been an interesting challenge.

Come to think of it, they may have already done this. How better to disguise an AI posing as a photographer than by having that AI propose such a plan?

[dsmithhfx]

Seems to have a bit of a fetish for mountains.

For whatever reason, that was a constraint set up by the programmers, so you could say it was their "fetish". Presumably it could be made to work with other photographic subject genres. Here's their precis:

Machine-learning excels in many areas with well-defined goals. However, a clear goal is usually not available in art forms, such as photography. The success of a photograph is measured by its aesthetic value, a very subjective concept. This adds to the challenge for a machine learning approach.
We introduce Creatism, a deep-learning system for artistic content creation. In our system, we break down aesthetics into multiple aspects, each can be learned individually from a shared dataset of professional examples. Each aspect corresponds to an image operation that can be optimized efficiently. A novel editing tool, dramatic mask, is introduced as one operation that improves dramatic lighting for a photo. Our training does not require a dataset with before/after image pairs, or any additional labels to indicate different aspects in aesthetics.
Using our system, we mimic the workflow of a landscape photographer, from framing for the best composition to carrying out various post-processing operations. The environment for our virtual photographer is simulated by a collection of panorama images from Google Street View. We design a "Turing-test"-like experiment to objectively measure quality of its creations, where professional photographers rate a mixture of photographs from different sources blindly. Experiments show that a portion of our robot's creation can be confused with professional work.

[emphasis added]

[dsmithhfx]

Maybe task it to join a few forums to argue about what brand of cameras are the best

Hilarious!

[pentageek]

Maybe task it to join a few forums to argue about what brand of cameras are the best.

... and catch LBA along the way

[RobA_Oz]

This has some of the hallmarks of design by focus groups. You'll inevitably get same-ish outcomes, because participants fit consumer profiles and mostly only like what they have seen before, so creativity tends to be stifled. Unconsciously ironic that they call this "Creatism", given that.

[ChrisPlatt]

Perfect viewing while you wait to be turned into Soylent Green.

Chris