Re:Style2Paints: AI colorization of images
by bencollier49 on Saturday September 30, 2017 @05:41AM
(#15372053)
I was thinking of trying it on something less Anime, and then noticed that another poster didn't like the original poster's subject-matter, so I ran it on this:
Re:Style2Paints: AI colorization of images
by anon203597 on Saturday September 30, 2017 @06:35AM
(#15372199)
Similar work on art style transfer usually has yielded somewhat surreal shifting results when they applied it to videos even after adding some logic that stabilized things from frame to frame.
Plus this is far from perfect, it works best if you provide a lot of hinting and even then it produces results that may look pleasing enough but are not really close to what one has intended for a particular character design.
I would guess that it can be made good enough for some novelty videos but it's still far from professional production quality.
Re:Style2Paints: AI colorization of images
by avian on Saturday September 30, 2017 @12:55PM
(#15373755)
I tried Style2Paints on a simple animation I did a while back. No manual color hints. I just played a bit with the references and settings in the web interface and used the provided models. The input were uncolored outlines. Considering it has very little in common with the anime training set I think it looks promising.
Re:Style2Paints: AI colorization of images
by julius on Saturday September 30, 2017 @06:41AM
(#15372221)
Interesting. It would be amazing if you could find the time, to elaborate, share some insights into this industry.
Specifically:
What are 2-3 major companies that would profit?
How does the current process work; do they make an entire uncolored scene and then color it?
What do I have to google, to find example videos of uncolored animation?
Can you ballpark how much it costs to color a 30min TV-Episode?
This depends a lot on the production process used. These days it's anything between hand drawn frames with digital coloring to completely digital processes using 2D vector graphics or 3D models.
Re:Style2Paints: AI colorization of images
by mistercow on Saturday September 30, 2017 @03:03AM
(#15371655)
I think there's a legitimate point to be made that it's a shame that the best way we've found to coordinate our resources to advance science and technology is capitalism. It's not like capitalism doesn't work, but if you envision a sort of "coordination oracle" which divides up resources ideally, and which everyone magically trusts, it seems pretty obvious that we would be advancing much more quickly.
But I also agree that there's not much point in dwelling on that point. So far, finding ways to leverage capitalism into solving problems has tended to be a lot more fruitful than looking for a replacement.
Re:Style2Paints: AI colorization of images
by Jack000 on Saturday September 30, 2017 @03:22AM
(#15371708)
Capitalism = greedy optimization with a Bayesian assumption. It tends to get stuck in local maxima and does weird things when the inputs are biased, but otherwise works ok.
There are almost certainly better ways to search the solution space, but most serious attempts attempts thus far have been non-sensical or impractical.
Re:Style2Paints: AI colorization of images
by Houshalter on Saturday September 30, 2017 @07:39AM
(#15372379)
No one objects that capitalism isn't a good optimization processes. Just that it's optimizing the wrong utility function. At one point something like 50% of STEM majors at MIT went to work on wall street. And the ones that don't go there go to work making more efficient targeted advertisements or addictive smartphone games or other unethical shit.
Our population has a limited number of "smart people". I don't think any alien looking in from the outside, would think we are allocating them even remotely efficiently. It would be pretty difficult to design a system that does worse than ours on this aspect.
The solution isn't necessarily communism, but perhaps a hybrid. E.g. the government funds things capitalism has no incentive to optimize for, like scientific research.
Re:Style2Paints: AI colorization of images
by KGIII on Saturday September 30, 2017 @11:29AM
(#15373289)
They may be smart, but they are still humans. You can't ethically assign them to career paths based on some vision of social good. That goes against the whole free agency thing and I'm pretty sure we've decided it's a horrible way to treat people.
That's well over the line into dictator territory. The results never match the predictions in the brochure. That's pretty much a sure fire way to get a tribunal convened at the Hague or a not-friendly visit from a SEAL team.
Re:Style2Paints: AI colorization of images
by Houshalter on Saturday September 30, 2017 @06:42PM
(#15375357)
Capitalism already assigns people career paths. How is that any more ethical? Not many kids grow up saying "I want to work on Wall Street" or "I want to make better targeted advertising." They go into those careers because they offer piles of money, and the jobs they might want to do don't. We could fix that.
Re:Style2Paints: AI colorization of images
by KGIII on Saturday September 30, 2017 @06:45PM
(#15375374)
No, it doesn't assign them, they choose them. Many people don't decide on a career path based on the greatest salary potential, but on things they enjoy, problems that interest them, and places they want to be.
Re:Style2Paints: AI colorization of images
by Houshalter on Saturday September 30, 2017 @07:37PM
(#15375547)
I think empirically people do choose jobs because of money. Otherwise why would Wall Street firms pay so well to attract talent? Why would people go to work there if they didn't care about money?
Re:Style2Paints: AI colorization of images
by KGIII on Saturday September 30, 2017 @07:43PM
(#15375583)
That's a very small number of people, not all of whom are well paid. The work is intellectually stimulating and has the ability to have far-reaching impacts.
I think you're confusing empirically for being significant numerically. I will offer a counter for your Wall Street example, look at how many go into teaching.
Re:Style2Paints: AI colorization of images
by Houshalter on Saturday September 30, 2017 @08:03PM
(#15375659)
>That's a very small number of people, not all of whom are well paid.
By 2006 about a quarter of MITs entire graduating class went to Wall Street. Some firms were offering $400K salaries to junior level positions. They weren't just taking huge numbers of people, but the ones they were taking were the best of the best. Other companies at the time had a terrible time recruiting anyone because they couldn't compete with that kind of money.
The problem with this is I don't think these firms are actually doing anything productive for society. The amount of work, money, and resources that go into getting a nanosecond edge for high frequency trading is just obscene. And all of the stuff they develop must be kept top secret and under NDAs. So even if they invent something cool by chance, they can't share it with the world or let anyone else benefit from it.
Re:Style2Paints: AI colorization of images
by KGIII on Saturday September 30, 2017 @08:44PM
(#15375804)
Yes, there are many more smart people than just those st MIT. I hold a Ph.D. (Applied Mathematics) from MIT. I've never even been to Wall Street, even as a tourist. I worked in traffic modeling, because the domain had interesting problems to solve and the opportunity to leave a mark was there.
Wall St. doesn't even employ that many people, in total. I'd also be curious as to how many remained working there.
Now, to remind us of the subject, we're discussing how it is immoral to determine career paths for other people. To be clear, I have no issues with providing incentives to get people to select a career. I'm good with that. However, demanding or forcing is right out.
People incentivized are still exercising free will when they choose to do so. Capitalism isn't determining their career paths. You could offer $75/hour to people to eat horse poop and you're still not going to get a lot of takers. People choose their careers for many reasons other than expected salary.
If they chose with salary as the criteria, we'd have no teachers. We'd have no researchers. We'd have no field scientists. Nobody would work for the government. We'd have few artists, few musicians, few authors.
People choose their career for many reasons, salary sometimes isn't even in the top three reasons. Of course, some people do so for monetary reasons. I'm not sure they are actually the most suited for the job.
The next time you go to get emergency medical treatment, do you want someone who is there for the paycheck, or do you want someone who is there because they want to save lives?
Now, your final paragraph... You don't think they are doing anything productive for society. I'm not sure why that's a problem. We are not the arbiters. There are very few jobs that someone won't point to and say the people performing those jobs aren't productive. Usually, they are Ayn Rand fans but I'll assume you're not one.
They create wealth. They create a system that enables transfer of wealth. They enable companies to be partially owned by the public. They increase the incomes of those who employ them. They enable people to retire comfortably. They enable employees by getting them greater value for their share of the company.
In short, they do lots of things that are considered productive. If I made you $10,000 in profit per day, you'd probably think I was pretty productive. I know the reverse is true.
Anyhow, your last sentence kind of draws a more complete picture for me. First, you speak of insisting people do certain kinds of work which you decide the priority of. That struck me as misguided, but I was willing to see it through. Now, you're lamenting that you're not being given the output of someone else's investment.
You can't ethically own humans. Insisting they do only certain tasks and then wanting the output of their labor is, well, treating humans as if the are your property. That is not okay, at least according to my morals.
If I'm misreading you, feel free to explain. However, it looks like you initially wanted to make people, smart people, do certain kinds of work. You've since added that you think paying them above average is like forcing them, that they are equal. You're now asserting that their labor output shouldn't belong to them.
Re:Style2Paints: AI colorization of images
by Jack000 on Saturday September 30, 2017 @08:49AM
(#15372623)
seems like we're basically in agreement. STEM majors working on ads are stuck at a local maxima.
On the other hand, humans are bad at predicting the future, so attempting to allocate resources perfectly is probably a waste of resources. see GPUs for video games, now used for deep learning - things that look trivial now might be important later on. If clicking ads gets us closer to AGI, maybe it's not so bad in the end.
Re:Style2Paints: AI colorization of images
by Houshalter on Saturday September 30, 2017 @09:53AM
(#15372855)
It's not that it's a local maxima. There really isn't any economic incentive to care about investing in technologies that can't be turned into a profit in a few years. Capitalism is doing a fine job optimizing, it's just optimizing the wrong function.
>humans are bad at predicting the future, so attempting to allocate resources perfectly is probably a waste of resources.
It's hard to do worse than "random" or "not at all", which is basically what our current system is. Sure it brought us GPUs. But that's pure coincidence that what video games require happened to be what AI requires. If we weren't so lucky that two completely unrelated industries happened to require the same technology, we would never get AI.
Imagine tomorrow someone make a superior graphics chip. That's somehow highly specialized to 3d rendering in video games and does nothing else. Dropping all the general purpose computation and linear algebra stuff. Gamers would all switch to it and the market for GPGPUs would die and AI would stop advancing, or even reverse.
And GPUs aren't that weird or surprising of an invention to invest in. Increased computing power benefits many scientific fields besides AI and would be an obvious investment for a hypothetical central planner.
Re:Style2Paints: AI colorization of images
by KGIII on Saturday September 30, 2017 @11:22AM
(#15373250)
We've tried the 'coordination oracle' a few times. It hasn't worked out very well at the larger scales, and I'm not sure trust was the issue.
It appears that it works better when the two systems are blended and neither is used to an extreme. It's really quite a broad spectrum between unfettered capitalism and central management of resources by the State.
Re:Style2Paints: AI colorization of images
by edanm on Saturday September 30, 2017 @06:11AM
(#15372140)
Well, if you had a "coordination oracle", you could probably convert most of the world's people to working on science and technology to advance the human race, and not spending any time on things like sports or fashion (or art). I mean, just cut 5% of the budget the world spends on fashion and use it to pay for science, and we'd have amazing advances.
The problem with this is that most people don't want to do this or think this is ethical in any way. (I'm more conflicted... you could take 5% of what the world spends on fashion to lift everyone out of poverty, too, as another example).
Re:Style2Paints: AI colorization of images
by NegatioN on Saturday September 30, 2017 @06:27AM
(#15372180)
I think this is more a labor of love than an attempt to appeal to the mainstream software community. The authors are also Chinese, where norms are somewhat different than the west.
Re:Style2Paints: AI colorization of images
by amelius on Saturday September 30, 2017 @05:54AM
(#15372080)
I suppose the images look creepy because of the combination of young age with poses and looks that are not usually associated with girls of such young age.
Re:Style2Paints: AI colorization of images
by anon203597 on Saturday September 30, 2017 @06:26AM
(#15372179)
Yes, let's apply cultural standards universally, surely this will make the world a better place. At least professional artists know how to avoid offending anyone's tastes.
Re:Style2Paints: AI colorization of images
by amelius on Saturday September 30, 2017 @08:51AM
(#15372635)
Perhaps the painting you referenced was meant to evoke disgust, or feelings of guilty pleasure. Imho, both are not really appropriate on a project page especially if used excessively and without reason.
At least the child in your painting looks naturally shy, given the situation. This in contrast to the cartoons, where the child seems to be taking the lead in seduction. Why ascribe feelings to children which they don't have?
Re:Style2Paints: AI colorization of images
by mrmondo on Saturday September 30, 2017 @08:09AM
(#15372491)
The painting you have used as a reference point of cultural standards is not a current style of art that is widely practised or recognised as ‘normal’.
Re:Style2Paints: AI colorization of images
by icebraining on Saturday September 30, 2017 @07:10AM
(#15372287)
What's absurd is determining the age in the first place. It's hard enough to do it reliably by looking at photos of Japanese Idols, let alone of heavily stylized characters.
Re:Style2Paints: AI colorization of images
by mrmondo on Saturday September 30, 2017 @07:57AM
(#15372451)
Indeed, they seem somewhat hypersexualised for their suggested age or at least hypersensualised to some extent.
Furthermore, and this is purely my opinion of course - they seem a bit... tacky(?) and I must admit if I had opened this link at work I would have been a bit embarrassed if someone cartoons of young girls like this on my display.
Perhaps at least a more diverse range of sketches would be both more relatable and useful to demonstrate the project.
Regardless, I’m not sure what rules of HN I broke to deserve -5 votes for voicing my reaction to a projects landing page.
https://imgur.com/a/z00jG
Not bad? I guess?
On the other hand: https://i.imgur.com/QkFDOVs.png
Edit: Also, i missed this earlier, but this video shows nicely how the AI can be guided with color hints: https://www.bilibili.com/video/av14443094/
If so, you will certainly have a market for this. The coloring work in animation take a lot of time and is very expensive.
Plus this is far from perfect, it works best if you provide a lot of hinting and even then it produces results that may look pleasing enough but are not really close to what one has intended for a particular character design.
I would guess that it can be made good enough for some novelty videos but it's still far from professional production quality.
https://www.youtube.com/watch?v=PZixy1cXQPE
Specifically: What are 2-3 major companies that would profit? How does the current process work; do they make an entire uncolored scene and then color it? What do I have to google, to find example videos of uncolored animation? Can you ballpark how much it costs to color a 30min TV-Episode?
key animation, genga (原画)
uncolored: https://sakugabooru.com/post/show/23064
finished: https://sakugabooru.com/post/show/23126
> How does the current process work;
This depends a lot on the production process used. These days it's anything between hand drawn frames with digital coloring to completely digital processes using 2D vector graphics or 3D models.
https://imgur.com/a/jXIhx
https://imgur.com/a/KJfMs
Who do you think made TensorFlow? Hint: it was Google.
But I also agree that there's not much point in dwelling on that point. So far, finding ways to leverage capitalism into solving problems has tended to be a lot more fruitful than looking for a replacement.
There are almost certainly better ways to search the solution space, but most serious attempts attempts thus far have been non-sensical or impractical.
Our population has a limited number of "smart people". I don't think any alien looking in from the outside, would think we are allocating them even remotely efficiently. It would be pretty difficult to design a system that does worse than ours on this aspect.
The solution isn't necessarily communism, but perhaps a hybrid. E.g. the government funds things capitalism has no incentive to optimize for, like scientific research.
That's well over the line into dictator territory. The results never match the predictions in the brochure. That's pretty much a sure fire way to get a tribunal convened at the Hague or a not-friendly visit from a SEAL team.
We are humans, not unthinking machines.
I think you're confusing empirically for being significant numerically. I will offer a counter for your Wall Street example, look at how many go into teaching.
By 2006 about a quarter of MITs entire graduating class went to Wall Street. Some firms were offering $400K salaries to junior level positions. They weren't just taking huge numbers of people, but the ones they were taking were the best of the best. Other companies at the time had a terrible time recruiting anyone because they couldn't compete with that kind of money.
The problem with this is I don't think these firms are actually doing anything productive for society. The amount of work, money, and resources that go into getting a nanosecond edge for high frequency trading is just obscene. And all of the stuff they develop must be kept top secret and under NDAs. So even if they invent something cool by chance, they can't share it with the world or let anyone else benefit from it.
Wall St. doesn't even employ that many people, in total. I'd also be curious as to how many remained working there.
Now, to remind us of the subject, we're discussing how it is immoral to determine career paths for other people. To be clear, I have no issues with providing incentives to get people to select a career. I'm good with that. However, demanding or forcing is right out.
People incentivized are still exercising free will when they choose to do so. Capitalism isn't determining their career paths. You could offer $75/hour to people to eat horse poop and you're still not going to get a lot of takers. People choose their careers for many reasons other than expected salary.
If they chose with salary as the criteria, we'd have no teachers. We'd have no researchers. We'd have no field scientists. Nobody would work for the government. We'd have few artists, few musicians, few authors.
People choose their career for many reasons, salary sometimes isn't even in the top three reasons. Of course, some people do so for monetary reasons. I'm not sure they are actually the most suited for the job.
The next time you go to get emergency medical treatment, do you want someone who is there for the paycheck, or do you want someone who is there because they want to save lives?
Now, your final paragraph... You don't think they are doing anything productive for society. I'm not sure why that's a problem. We are not the arbiters. There are very few jobs that someone won't point to and say the people performing those jobs aren't productive. Usually, they are Ayn Rand fans but I'll assume you're not one.
They create wealth. They create a system that enables transfer of wealth. They enable companies to be partially owned by the public. They increase the incomes of those who employ them. They enable people to retire comfortably. They enable employees by getting them greater value for their share of the company.
In short, they do lots of things that are considered productive. If I made you $10,000 in profit per day, you'd probably think I was pretty productive. I know the reverse is true.
Anyhow, your last sentence kind of draws a more complete picture for me. First, you speak of insisting people do certain kinds of work which you decide the priority of. That struck me as misguided, but I was willing to see it through. Now, you're lamenting that you're not being given the output of someone else's investment.
You can't ethically own humans. Insisting they do only certain tasks and then wanting the output of their labor is, well, treating humans as if the are your property. That is not okay, at least according to my morals.
If I'm misreading you, feel free to explain. However, it looks like you initially wanted to make people, smart people, do certain kinds of work. You've since added that you think paying them above average is like forcing them, that they are equal. You're now asserting that their labor output shouldn't belong to them.
Err... That sounds remarkably like slavery.
On the other hand, humans are bad at predicting the future, so attempting to allocate resources perfectly is probably a waste of resources. see GPUs for video games, now used for deep learning - things that look trivial now might be important later on. If clicking ads gets us closer to AGI, maybe it's not so bad in the end.
>humans are bad at predicting the future, so attempting to allocate resources perfectly is probably a waste of resources.
It's hard to do worse than "random" or "not at all", which is basically what our current system is. Sure it brought us GPUs. But that's pure coincidence that what video games require happened to be what AI requires. If we weren't so lucky that two completely unrelated industries happened to require the same technology, we would never get AI.
Imagine tomorrow someone make a superior graphics chip. That's somehow highly specialized to 3d rendering in video games and does nothing else. Dropping all the general purpose computation and linear algebra stuff. Gamers would all switch to it and the market for GPGPUs would die and AI would stop advancing, or even reverse.
And GPUs aren't that weird or surprising of an invention to invest in. Increased computing power benefits many scientific fields besides AI and would be an obvious investment for a hypothetical central planner.
Except not everyone agrees on what is ideal, and therefore such an oracle is impossible even as a limit to be approached.
It appears that it works better when the two systems are blended and neither is used to an extreme. It's really quite a broad spectrum between unfettered capitalism and central management of resources by the State.
The problem with this is that most people don't want to do this or think this is ethical in any way. (I'm more conflicted... you could take 5% of what the world spends on fashion to lift everyone out of poverty, too, as another example).
https://commons.wikimedia.org/wiki/File:William-Adolphe_Boug...
At least the child in your painting looks naturally shy, given the situation. This in contrast to the cartoons, where the child seems to be taking the lead in seduction. Why ascribe feelings to children which they don't have?
Furthermore, and this is purely my opinion of course - they seem a bit... tacky(?) and I must admit if I had opened this link at work I would have been a bit embarrassed if someone cartoons of young girls like this on my display.
Perhaps at least a more diverse range of sketches would be both more relatable and useful to demonstrate the project.
Regardless, I’m not sure what rules of HN I broke to deserve -5 votes for voicing my reaction to a projects landing page.
These are fictional characters, they can be 500 years old and have 3 penises.
You see, you can't tell the age, you can't tell the gender, because they are not real.