Typesetters on Suicide Watch

Where were you when our robot overlords put every typesetter and cleaner out of business?
imgur.com/a/SVkOPCp

Attached: 2xu56YA.png (626x223, 124K)

Other urls found in this thread:

gelbooru.com/index.php?page=post&s=view&id=4894057
vice.com/en_us/article/4x3pwm/meet-the-scanlators
twitter.com/AnonBabble

Does this faggot release sources or the tool itself at least somewhere?

He's being yelled at by Leddit to patent his code now, and like a fucking moron he's still wanting to release it for free.

>gives google-translated text
lmao no thanks

All it takes is changing it to accept an actual script, which is pretty trivial.

Yeah, I'd say he shit on patents and release sources. He can still get hired for a mad sum at any big name company afterwards. Patenting code is not for cents, anyway, and only possible in burgerland.

Even the author mentions that in its current state, machine-translating Japanese is a no go. But finding and filling bubbles is already a big help in itself. Though it can't deal with text overlaid on art (yet?).

>google translate does not recognize words as simple as "gumi" as in "shinsengumi"

kek

>typesetters and cleaners already out of business
>one day machine translations will become as good or better than human translations
>eventually even redrawing will be automated
WHO'S LAUGHING NOW YOU SMUG CUNT?

Attached: 1559932865248.png (300x200, 137K)

The bubbles are always the easy part mate, this shit is almost worthless.

Not needing to coordinate a team to get translations out really helps streamline

i dont care about this gay shit. when will we get an AI that transforms my manga waifu into a 3d model?

Which part of almost worthless you don't understand? You still need someone to redraw images with text and typeset text there, bubbles were never a problem.

And yet 9/10 faggots fuck bubbles up with Wild Words, shitty line breaks, metric kerning, and stupidly big spacing.

What's your problem, faggot? Does it hurt you that parts of manga translations now can be automated? It always starts somewhere and will gradually improve. Don't tell me you're making a living selling your work to "alternative" translators? In that case, my apologies. No, not really.

we already have redrawing options

You'd still have to adjust things unless you're alright with some bubbles looking like shit.

it has limited uses but there are already quick ways to clean up bubbles and input text
a decent text shaper would be helpful though

I mean it's got only 4000 samples it learned from. I don't even have the slightest clue of how machine learning and AI functions but that's pretty impressive. If it gets even more samples to learn from and it's able to recognize different manga then that would be pretty good since a lot of work would be erased with that.

>one day machine translations will become as good or better than human translations
Most of us will be dead by that time. Learning nihongo is still the way to go.

I doubt it's all that far off actually. Mechanical translations are useless but neural networks and machine learning are going places.

Literally takes less than a year to learn to 3D model her yourself. And before any
>but a year is a long time
consider how many years you have already wasted not doing it. If you picked up a new hobby now, you'd be making a waifu a week before valentines.

Yeah, now try typesetting non-trivial things like SFX

Attached: [Commie] JoJo's Bizarre Adventure - Stardust Crusaders - 02 [921E71F7].mkv_snapshot_03.55_[2014 (1280x720, 83K)

not anytime soon. and the first languages to go will be the ones like french, spanish, german and the like. korean, chinese, japanese and english are all safe for quite a while.

I'll be pessimistic about this, but I guess it might be a bit over-fitting since he only train/test it on OPM. Yes, the test isn't in the train, but it is still the same manga, the similarity is there.

Not a bad start though, I could see this be generalized and greatly help in scanlation process.

>one day machine translations will become as good or better than human translations
Not before your timely death. Better learn it now.

Attached: 1468896682567.jpg (850x1200, 141K)

Idiot. His tool is for speech bubble detection, not for translation.

A lot of scanners just use google translate anyway

Attached: 1555585356535.gif (500x545, 263K)

>colors speech bubbles white and fits text in them
A literal monkey could do that part of the job. 95% of the effort goes to redrawing and fitting in text outside of bubbles. This isn't going to replace anyone just like how google translate isn't going to replace translators. Or anyone of value at least. Plenty of typesetting work out there already looks like it was lazily slapped together by a computer.

I would assume 4000 samples should have taught it to center text at the least.

>one day
Do you understand that the machine must understand the context to pull it off?
It's most likely will be the day when human civilisation ends.

What about mechanical translations + reading comprehension to know what's being said in the broken english you get + the ability to rephrase in fluent english?

Dumbest comment of this thread so far.

>wtf someone might have to proofread and make small manual tweaks after, that never ever happens with humans making mistakes, this is completely useless
bakas

Elaborate.

No one does typesetting because they want to, they do it because someone has to.

but scanlators sick mothers...

Hey, I at least check it against online dictionaries and use context to improve the translation.

Make a tool that auto levels pages and cleans image noise.

What are some of the better ones? I tried the option in photoshop and still thought the results weren't very good even after several attempts to use content aware fill.

this is not the first time we're seeing this. this comes up every now and then, newfags please.

Given how some translators don't seem to know Japanese that well to begin with, I think you underestimate how neural network translations could be better than that segment of people.

you can still use a dictionary and go through it word by word.

Imagine if the scanlation community of scanners, cleaners, typesetters, translators, editors, QA, etc, etc, etc, spent all of those hours learning to draw and tell good stories?
Western comics could become competitive again if they didn't devote all their obsessive energy to ripping off Japanese stuff.
But then again, I guess they did it to use with heavy industries and consumer electronics, so fair play.

you should not talk about things you do not know about.
go look at the state of western animation. it's not copying anyone and it hasn't produced anything equal to avatar the last airbender in the last 10 years.
the west seems to be dominating in movies and video games.

You can still do everything manually, yes, but that's not the point. Though any sensible translation must flow through a competent editor, at least, whether it was human or machine translated.

use it to clean the jap sound effects off your image if you're so confident in it.
robots will never replace people fully.

You should not make posts which, though short, are non sequitur. You should learn to connect multiple thoughts together and read the posts you are replying to.

Imagine replying to, not even bait, just an obvious joke to talk about something cool, with an insecure "n-nuh-uh the robots need us".
For fuck's sake, they're just computers.

>robots will never replace people fully
Why? I'm not saying it will happen, but in a sense humans are just organic robots as well, so we already know it's possible on some level. Do you think there's something about carbon-based biological computers that can't be replicated using silicon-based computers?

That's cool, I'm translating a doujinshi to help myself learn nip and am doing the typesetting right now for it and I wish I could just skip it all together it's boring as shit and a pain to get just right, it should absolutely be automated.

I don't understand how automating that is a "patentable" idea you can't patent the idea of using a neural network to detect something, it's much the same as neural networks that re-draw censored hentai and other stuff. Though the American patent system is completely broken.

>clean the jap sound effects off your image
Why do people do this? It's part of the original art work. At most typeset the translation next to the original sound effects like Ginrou translation team does.

If this is bait, well done, but if it's a serious question the answer is that neurons have built-in fuzzy logic due to the electrical-hormonal interfacing we use, so we can work out complex intuitive things better, pound for pound, than any binary circuits can. So far, at least, it may change with better logarithmic programming and machine learning.

Because it's a meaningless part of the original unless you already understand Japanese, in which case don't read scanlations, and it's usually more typesetting than actual art anyways.

Replacing it often alters the original art work though, since you have to use content aware fill to guess what's underneath it because the English typeset won't cover it completely.

Then what would you do about narration or background dialog that's outside of a bubble and covers the art?

>Though the American patent system is completely broken.
This kind of stuff is always patent-able. Of course, you cannot patent the "using neural network to detect something" since it is too fucking broad, but you definitely have better chance if you phase it a bit more specific like "detection of text boundary of text in japanese manga".

>is that neurons have built-in fuzzy logic due to the electrical-hormonal interfacing we use
Not that user, but is this a speculation or something that's already proven? Proven as being observed in anything more complex than mice.

Slightly transparent dialogue box.

Attached: TextType.jpg (853x1200, 431K)

Stop trying to justify your laziness.

Never! I must fight for my right to be lazy.

Reddit is retarded, those sorts of patents blow over in a strong wind, and the correct answer would be to license and sell the source code to companies.
The main reasons you all have for calling the US patent system "broken" is anecdotes about people who know it exists but choose never to use it getting screwed, which, yeah. The system does not protect people who specifically ignore it, allowing others to impersonate their discoveries. No shit. Muh poor Tesla, all he ever had in life was a state of the art lab and access to a city-sized power plant handed to him by J.P. Morgan, what a shame that bastard Bell actually took a second to file paperwork while Tesla spent years moping around not even having the courtesy to marry Morgan's daughter, which is the main reason he gave him the lab in the first place, and having no interest in intellectual property or money anyways because he was just a really smart NEET with a sugar daddy.

it's an oldschool scanlator method

Attached: great-teacher-onizuka-92770.jpg (800x1272, 172K)

Wow. Even I'm not this lazy.

What?
Has...has the observable function of neural networks been...proven?
Why on something "more complex than mice"? Why would you ever even ask that?
Do you think that mice would have miracle neurons that humans don't, and that humans do not in fact have hormones involved in their brains and nervous systems?
I simply don't understand what this comment is even trying to ask, do you know what brains are to begin with?

it's color coded for each character. feels like he put effort into the wrong places.

This gives me the perfect idea for MangaDex's next April Fools Day window to post joke chapters.

Waifu2x?

Isn't that mostly for upscaling low resolution scans? I toyed around with the denoising feature but wasn't really impressed with the results. I didn't think it auto levels.

it's been done so many times already.

It's mostly for removing artifacts on color pictures, it does kind of an incredible job at it.

>and like a fucking moron he's still wanting to release it for free.
Yeah what a moron for wanting to help humankind without profit. What a fucking moron he is for not being a greedy piece of shit.

First manga fully done by an AI when?

Attached: lain4.gif (972x1000, 1.09M)

Yes. If he were really such a martyr, he'd sell it for maximum profit to someone who would then produce the translations for dirt cheap and sell in volume, and he'd use his proceeds to, you know, donate to the WHO or something more important than giving your leeching NEET ass free manga. That's just a waste of resources since you yourself will never do anything for your species in return.

Would there be a public for a manga site where you hover the mouse to see the translation? Like the boorus but without the annoying rectangles

gelbooru.com/index.php?page=post&s=view&id=4894057

If only there were some way to download these with the image, it would be great, as is it's pure shittu.

Yea he is a fucking retard. Just think of all the scanlators who do free work and still get nothing but harassment from entilted turds, meanwhile the people who enjoy their free labor have no good words to say because they take it as a given that something be translated. People who do free work are the kind of people who get stepped on until they're broken

imagine still not knowing nip
imagine

Will we ever get proper scans of GTO?

Not to mention all the mangaka who don't even bother with localization because they know literally everyone will just read the scanlation free online instead, and if they put up an online official translation for money it'll just be stripped and reuploaded on a thousand sites immediately, and if they put it up just with ads for a few pennies, people will just use adblock and/or put it on scan websites with familiar interfaces that they like.
The mangaka is really a masochist, making their product for their own little market to be enjoyed free world-wide, pushing each other out of the limited slots in weekly publications, burning out just trying to get kids to enjoy their art.

Speaking of gelbooru, do they use machine translation of some kind, or do people contribute translations? Seems kind of weird not to typeset them if it's being translated by a person.

Typesetting would require even more image families and they already have a billion images, and it's FAR less work to type the translation and stretch the box over the original uploaded image than it is to download it, edit it, clean it, typeset it, and reupload it, then go through whatever request process gets them recognized as alternates in a set, or related image variants, so people can click various links around to finally get to the one with the translation, all for someone to bitch about the font you used.

So by doing it this way they get 10x more translations from random fappers than they otherwise would, because it's low-effort. Also it's not downloadable so they can serve you ads every time you want to see it. Boorus are generally aggregators, not creation nexii.

Kind of like what's going on with Citrus+ right now. The manga has only been officially released in Japan, but because the author is busy preparing for the official international launch, she's been too busy to release new chapters the last few months. Really, only the Japanese fanbase should be annoyed, but people who only read the free fantranslations of her works are really annoyed right now because they don't get anything new until she has time.

Yeah, it's pathetic honestly, but try telling Yea Forums that, you're just a corporate/Jewish shill, with the emphasis being which side of the political autism spectrum they fall on. It's so UNJUST that art isn't a human right! Free the information! A product only has value if it's physical, yet it's vitally important that we all be able to get these valueless products people pour their lives into! Buying manga bad, scanlation good, stealing other people's work bad (but only scanlator to scanlator, not including the mangaka!).
Boy, I wonder why manga keeps catering more and more to Japanese normalfags, couldn't possibly be that mangaka want to eat! Don't buy, you're supporting evil distribution companies who make a profit by producing the physical products and handling marketing and exposure, which isn't "real"!

The Hellsing and Drifters mangaka said he hopes people who pirate his manga get a horrible disease. So he wants pirates to suffer and possibly die.

vice.com/en_us/article/4x3pwm/meet-the-scanlators

Based.

THEY TOOK OUR JEBS

Attached: angry mama.jpg (1240x1080, 457K)

>the estimated cost of damage from online piracy, they said, is approximately $20 billion dollars.
Imagine being this delusional

Attached: 1562151111683.png (200x200, 48K)

oh fuck shitposter-kun just got called out in the stone thread, which of you fags was it?

>if I post le edgy supernatural whore smirking with my image it means my opinion is right and economic projections don't real
Sub-90 IQ.

Measuring loses of an international market that historically grew through piracy is fucking stupid. Measuring loses through downloads is stupid. you're stupid.

>the future is here

Attached: 1435116307946.jpg (764x744, 220K)

You're sub-90 IQ.

>machine translations better than human translations
>ever
Man, it's a good thing you're reading simplistic children's cartoons instead of anything that's actually difficult. Human beings can't even meaningfully translate across human languages. Everyone who actually reads understands that translated versions of high literature are basically pointless, because you lose everything that makes it "high literature". There will never be a better machine translation until we've got actual AI instead of just banks of IF using statistical models, and at that point, you'll still be dealing with a third-party translating something twice removed from the original source-encoder-recipient relationship.
The only way machine translations will ever be meaningful is if we all have the same culture and ideology, and we'll almost certainly have to all speak the same language to get there. The majority of people will be speaking some pidgin of English, Chinese, Spanish, and French at roughly the same time something like AI will actually be able to exist, meaning it'll be pointless.

And all of that is happening either after you die, or decades from now. You should probably just learn the simplistic nip you need to know to read your children's stories instead of pretending it's impossible to do it at all. Military linguists learn Chinese in less than a year, and no one who is smart joins the military. You've got no excuse other than laziness and a lack of dedication.

>Immediately after a Berserk chapter's release, he and his team of five scanlators begin to slave hours away over Photoshop with a Japanese dictionary. Scanlating one manga chapter can take them over 30 hours.

Should it take this long? They don't even do redrawing do they?

>dude just sell it to chinks that will just throw in the market billions of mangas machine translated

What a great idea user, i'm sure this will only increase manga sales overseas

>things retards who never scanlated anything for free say
Nobody does this. In fact the vast majority of readers will defend any shit translation as long as it's free.

The other issue is context. Many words have more than one meaning and you have to be looking directly at the pages to be able to tell what meaning it is.

>muh trillions of lost sales
it's a fucking vice article, leave it to the redditors ITT to unironically take it seriously

No, there are plenty of whiners who will complain and something was "mistranslated" because it wasn't translated the exact way they felt it should be despite the meaning being basically the same.

>plenty
No, not really. Scanlate something for free, then come back and say it again. That happens once in a blue moon and when it does it gets instantly silenced by other people defending it because it's free.

I must find and kill him before he releases the demon.

Attached: scanlation crimes.png (424x288, 62K)

What the fuck do you think "actual AI" would be, codelet?

What kind of magic is this

Attached: tejina excited.jpg (945x1079, 118K)

>still staring at pages
The context is even larger than the medium or something as simple as the difference between connotative and denotative language. How do you translate unreliable narrator, something that is almost always lost across cultural boundaries right now with human translators who understand the culture and history? How do you even begin to effectively translate "Point deer, say horse," and still effectively communicate it without a history lesson in Chinese politics? You can't.

>le computational thinking is king
It's not a stack of IF shoved around with babies-first linear algebra lesson, mathlet. Computers are stupid. They cannot be intelligent, because they do not have intelligence. No amount of being able to differentiate between objects based on statistical models is going to give a computer the ability to define new information unless a human being goes in first to tell it how to do it.

When whatever the architecture looks like can teach itself how to define things without a human being doing it, you'll have AI. We are not even remotely close to anything like that just because black box code happens out of genetic algorithms and compsci monkeys are too stupid to realize that's the fucking point of a genetic algorithm as defined 30 goddamned years ago.

I swear I've seen various versions of automated typesetting posted a bunch of times, and people regularly use them, but none of them ever seem to revolutionize anything or be much else than a minor time-saving tool, so I don't think this one will either.

>How do you even begin to effectively translate "Point deer, say horse," and still effectively communicate it without a history lesson in Chinese politics? You can't.
You localize the line by using an English expression that means the same thing.

Says the idiot who's only response is "sun 90 iq!!!" Shut the fuck up you stupid kid.

There is no localization of "Point deer, say horse," because the phrase in Chinese is completely meaningless unless you know the historical context to it. It does not have a localized equivalent anywhere outside of China, because the phrase refers to a singular, specific event in Chinese political history that has been transformed into a lesson about politics that people invoke by saying a one-liner.

You're basically saying it's possible to localize an inside joke to people who aren't in on the joke. You can't do that unless you explain the joke.

Having to explain the entire joke makes the whole thing lack the intended impact. In that case you might as well ignore the whole idiom and call the person it's referencing deceptive instead.

>all this translators in full damage control
LMAO

Nigger are you actually dumb enough to be seeing computers as a threat to you and arguing about what "real thought" is because you saw too many sci-fi movies and know actually zero about coding or processing?
God, I wish you fucking retards would just never talk about computers, or brains, if you don't know the first thing about them. So fucking stupid, actually infuriating to have to keep hearing this emotional, religious drivel brought into matters of simple manufacturing and labor by superstitious zealots.

god, this post is such an unfortunately clear view into the ugly twistings of brainlet consciousness

I for one embrace any tool that might make my machine translations take less time to do.

Fuck off, [insert popular isekai translation group]

BEAST'S LAIR ON SUICIDE WATCH
is that place stll alive? can I even tell the to kill themself?
truly the hero of justice we need

According to Mangadex that group had new content posted just late July.

Military linguists also go to schools that take up a vast majority of their waking hours, a far cry from people who actually have day jobs

Honestly, I think this kind of stuff is a net negative.

Right now, the barrier to entry is time. Most people are not going to do all the cleaning work and shit if they don't at least have an actual translator. With this kind of software, literally any retard can translate manga, and with no real quality control.

The shittiest part is that bots auto-upload it to manga reader sites & more than a month can pass before it gets fixed.

Happy trolling.

Wow, it literally does the easiest part of scanlating.

There are already photoshop scripts that help erase speech bubbles. And really, besides a few basic rules, like center the damn text and use a decent all caps font, any moron can already typeset to a certain degree of quality.

And there are already plenty of groups that use edited machine translations.

>He's being yelled at by Leddit to patent his code now, and like a fucking moron he's still wanting to release it for free.

What's the fucking point or patenting it & what exactly is he going to patent in the first place? It's not some advanced algorithm but a fucking trained neural net.
What's next? Copyrighting every single function in a program? Beyond retarded.
Also, this is like a spit in the ocean as far as scanlation scene goes. Trading two clicks for one, WOW SAVED THIRD OF A SECOND.

That small time saved can really add up,especially when you're working on several chapters at once.

>Copyrighting every single function in a program?
You do know you live in a world where Apple and Samsung sued each other over the definition of what a "Tap" was, right?

>Trading two clicks for one, WOW SAVED THIRD OF A SECOND.
>Saving 50% of an actions required
>Saving 33% of the time needed

You are absolutely retarded and your opinion is worth shit.

>Saving 50% of an actions required
>Saving 33% of the time needed
What the actual fuck.
Back to school.

Attached: 1567459200659.jpg (414x508, 27K)

>here, thanks to machine, no need to do this anymore
>here, thanks to machine, no need to that either
>here, thanks to machine, no need to learn this anymore
>here, thanks to machine, you can save time
Am I the only one who's not pleased by all of this shit? Soon, machines will be in charge of everything: translation, music, construction, maintenance of other machinery even. And humans will wallow in self-indulgence and do nothing with their lives and just live in order to wait for the time of their death.
This is saddening, really.

Attached: sad yuki.jpg (1498x1022, 159K)

>humans will wallow in self-indulgence and do nothing with their lives
They'll create a virtual reality where they can simulate all the struggles and duties that their lives used to have in the early 21st century. They'll call it the Matrix.

There is nothing to patent here. Its all basic bitch shit based on mathematics everyone has used for 100 years. The problem is getting training data.

I just found a computer program that can do 90% of my cleaning work for me! All I have to do is darken the blacks a bit and do a few touch ups.

you're an absolute clown, a robot will never be able to accurately translate something like
ワロスwwwwwwwwww
or
マクドの場所ぐらいggrks

Attached: 考え込む.png (283x294, 89K)

>implying
Of course it can.

Technically you could program in recognition of certain slang sayings like that. The bigger issues are context of the scene in question and figurative sayings and idioms which won't make sense when translated literally

Another issue is that Japanese can be very terse and leave key parts of sentences out that you would then have to figure out from the surrounding sentences. A machine translation tool is very far away from being able to do this. Add in the fact that words can have multiple meanings and you have to look at the manga panels to figure out which meaning it is.

Machine translation will for the foreseeable future require human oversight.

The Chinese course is 72 weeks

>you would then have to figure out from the surrounding sentences
Not really, since manga has visual representation also.

Which would be a problem for a machine, since it doesn't have eyes. So a machine will require a someone to oversee it.

Finally I see