[WIP - Full] Virgin's Protection Magic Translation (v 0.82 released 23/07/2017)

[Renestrae]

The name JapeTrans isn't going to change, just the name I use when publishing it. This isn't a 100% guarantee of severing the link, but that's fine. The most likely scenario is that someone looking at my CV sees it and clicks a link to verify that the project exists, because they've got multiple CVs to consider and a deadline. Even if they do search for it, they're not going to click on a link to somewhere like ULMF because they're at work and really shouldn't visit a porn site (imagine the justification the recruiter would give to their boss for turning me down. They'd probably get as far as "Well, I was searching around for info and I ended up browsing a porn site..." before being shown the door). The other alternative is that the person is already familiar with my work, in which case no amount of renaming/obfuscation is going to hide the link. But hey, at least in this scenario they'd know I can write pretty well.

Haha, fair enough. Consider the other way around too, maybe u don't want degenerates stumbling on your work GitHub? Eh, maybe I'm also a little over-paranoid about this whole separation deal

 

[u5ern4m3123]

I'm really glad to see this thread open again. I hadn't checked in so long that I had forgotten which site this thread was on, and was a bit concerned when I heard HF died recently. Not just for VPM, but also JapeTrans in general, which I'd really like to see get completed and widely used. Glad to see the thread alive again!

Haha, fair enough. Consider the other way around too, maybe u don't want degenerates stumbling on your work GitHub? Eh, maybe I'm also a little over-paranoid about this whole separation deal

As someone who knows a thing or two about programming and is impressed with Habisain's work with RPGMakerTrans, I was actually really curious about what other coding projects he's done. So, I hope he doesn't mind a few new GitHub followers from seemingly random people, as long as those accounts don't connect to any porn-related stuff.

On that note, I'm a bit curious if the decensoring research on his site went anywhere. It seems the last update on the page is from Jan 2018. I know there have been a few more recent efforts, but his was the first I came across, so I thought I'd ask.

 

[habisain]

On that note, I'm a bit curious if the decensoring research on his site went anywhere. It seems the last update on the page is from Jan 2018. I know there have been a few more recent efforts, but his was the first I came across, so I thought I'd ask.

Yeah, I need to update that site. Anyway, the answer is "sort of" - or at least, it went far enough for me to know that those ideas are not a viable end product. To a very large extent, those ideas have actually been independently implemented in rexkaXua's Demosaic Project (

You must be registered to see the links

), where they've used a combination of traditional computer vision based block detection and neural network upscaling based on low-res sprite upsamplers to implement an automatic demosaic (although they don't appear to be tuning the upsampler network, which they probably should). The problem is that it doesn't work very well quite a lot of the time, and something like 5% of the time it creates output which (I at least) find genuinely disturbing to look at i.e. a critiical failure. I'm sure some people get off on looking at drawings of deformed genitals, but not me. The problem is exacerbated by the fact that for understandable reasons, many Japanese hentai artists don't put much detail on genitals which are going to be pixelated in the first place, meaning that in many cases even if you had a perfect upscaler, you still wouldn't get an acceptable result.

I'm currently (albeit slowly) investigating an approach based on autoencoders, which I think is the "correct" way of doing this, or at least the best method using the current state of the art techniques in ML. If you've not heard of autoencoders, they're the technology behind DeepFakes. The current problem I need to solve is that I've ended up with a latent space which is far too big for the amount of training data available. So that's a genuine research problem that I'm looking at. But if the training input/latent space problem can be solved, an autoencoder based solution is capable of replacing the pixelated data with "equivalent" non-pixelated data, so it shouldn't have the same issues as just upscaling (it probably will have different issues though, especially w.r.t. drawing styles, but that's what style transfer networks are for...)

JapeTrans's code now lives on SourceForge and is easily findable by search. As a long-time Mercurial user (and someone who finds Git to be a deeply flawed tool) my hosting options don't include GitHub. Unfortunately I haven't got any other projects up there yet though (I probably will at some point, but waaay to much stuff going on right now.)

 

[u5ern4m3123]

I actually couldn't find it through SourceForge's built-in search bar, but Google worked. You'd think that's a pretty basic thing to get working for a website. I'd never actually used SourceForge before, as it always seemed less open and user-friendly than sites like GitHub, and the signup form requires info like location, job title, etc, which seems odd to me. At least it seems that info is kept private, as well as what projects I follow (which is nice).
I had figured you hit a dead-end with that initial idea, as I only noticed it referenced once with no update, but it's nice to hear that you are still looking into other methods. Data research and neural-network type-stuff is a bit out of my wheelhouse, but if you need a dataset, did you consider Danbooru2019? It seems to be everything from the Danbooru site, with tags and metadata, and is apparently so large that the next release will be too big for torrents. The site says it contains "~3tb of 3.69m images with 108m tag instances," so there's bound to be some stuff you can use to fill out your training data. The link is here for more info:

You must be registered to see the links

 

[habisain]

Sourceforge is a pretty nice place now. It's got a very long history (not all of it nice)... But odd how their built in search doesn't work as well as Googles. And yes, they ask for a fair amount of data, and no, they don't require any of it to be accurate. I think it's just because they hail from a different era of the Internet (i.e. 1999). In any case, for me it beats using Git, or trying to figure out how to convert my repositories to Git (which doesn't support all of the features of Mercurial, including some really quite useful stuff).

Regarding the decensoring stuff, I think we should take this to PMs if you want to continue this discussion - it's getting somewhat offtopic. I will say though that the Danbooru dataset is part of the problem in this area though, not part of the solution. It doesn't contain the necessary bits of data to train this kind of model (censored/decensored image pairs), and so it requires a lot of processing to be useful. I think Deeppomf did some of that processing, but hasn't released that dataset (understandable due to copyright etc), and I don't think they got particularly good results (on the grounds that the Deeppomf models aren't being used now). In any case, models trained from the Danbooru dataset tend to work by the upscaling method, which means they're answering a slightly different problem, and thus get a solution that's slightly not the right one.

Going to take a break from ULMF for now though, so I won't see any new posts in this thread for a while (although PMs will get through via email). It appears there are too many virulent rape apologists here, which is depressing, and I don't need to deal with that kind of stuff.

 

[habisain]

As it's probably overdue, here's a quick post on the status of this project

1) Why's it taking so long?

There were a few complications. Over 2019-2020 my run of good luck with computers thoroughly ended, and I've suffered 4 computer failures over that period, of varying severity. Obviously that slows things down. However, the bigger issue recently is the impact of the COVID-19 pandemic. Specifically, this project was something I primarily worked on during commutes, and thanks to COVID-19 I now work from home, so no more commutes. It took an embarrassing amount of time for me to realise the issue, so hopefully things will start progressing a bit faster now I'm actually allocating some time to work on this.

2) JapeTrans

JapeTrans continues. It's at about 90% done, excluding the GUI. It's also picked up a bunch of features that were low effort to implement but with outsized benefits. For example, it now has a binary patch format with cryptographic verification options, because fun fact: since RPGMaker XP, all RPGMaker engines can run arbitrary code on your computer, which means that any translation patch is a security risk (I'm aware that it's probably not that much more of a security risk than the pirated games a lot of people run, but still). Cryptographic verification at least produces something approaching a chain of trust, so that's an improvement, and the binary format has other benefits too (i.e. not having to parse the patch, more compact representation, compression). There's also support for multilingual patches (i.e. a patch that contains more than one target language, so in theory I could add in, I dunno, French or American English).

There's also a feature that's implemented for improving the lot of translators, but that's still being shaken out so I'll hold off on announcing that for now.

3) The VPM Patch

Is in fairly good shape, actually. It's advanced from where it was, certainly. One of the annoying things that did happen is that one of my computer failures exposed a problem in my backups, which was that while the patch was backed up, the version history of the patch was not, so I've lost a lot of the version history. However, that's probably not of too much interest to anyone other than me, because it means I don't have a good log of what I've done to it since the last release anymore.

 

[Darkstar816]

Thank you for sticking with VPM.

I really enjoyed what I have played up to now, and look forward to having a complete playthrough someday.

 

[BigJohnny]

The dev just announced a translation... to Chinese-

You must be registered to see the links

 

[Bryanis]

The dev just announced a translation... to Chinese-

You must be registered to see the links

And interestingly enough, the article is there twice... once in Chinesse and aother time in English.
Maybe a sign toward an English translation (or toward the westerner) ?

 

[habisain]

And interestingly enough, the article is there twice... once in Chinesse and aother time in English.
Maybe a sign toward an English translation (or toward the westerner) ?

Not as such. Searothonc can't write in Chinese, so he's written the article in English, and then machine translated it to Chinese. He's then posted the English version in case of Machine Translation fail.

Still, I might get in touch with him at some point and see if he's interested in making an official English version.

 

[Bryanis]

Wait to shoot any hope down ...
But better to face reality ^^

 

[Darkstar816]

Still, I might get in touch with him at some point and see if he's interested in making an official English version.

After what happened with Teaching Feeling, I am a bit worried about a repeat happening. Can there be some measure taken to ensure that the VPM translation isn't killed? An attempt at an official localization definitely should be done, but...I worry.

 

[Bryanis]

Well, teaching feeling is different since the fan translation was stopped at the demand of the dev since "he contracted another entity to do one for him".

Here if the dev tell to back of doing a fan translation, any fan can just do it on it's computer and spread it on internet. As long as he don't respect the dev wish and / or the dev don't bother hunting down the translation process / sharing.... That's what's happening for 99% of the fan translation. Dev either don't know, don't care, or can't bother to hunt down translator and sharer.

But considering that Habisain is mostly done with the translation and that it's easy to offer just a patch requiring to own the game. Seem like the dev has nothing to loose to let our friend finish than shutting him down...

 

[orangestone]

Still, I might get in touch with him at some point and see if he's interested in making an official English version.

Any progress with this?

 

[habisain]

Any progress with this?

I can't really say right now.

 

[Neosuduno]

Could you tell us how far along you are? We haven't had a progress report in a while.

 

[habisain]

Could you tell us how far along you are? We haven't had a progress report in a while.

OK, so the thing that I can say: I've had some health issues - If people haven't noticed, I haven't really been active here over the last few months, and that wasn't by choice. Basically I suffer from hypermobility, and I was advised by my doctor to reduce my computer usage for a while to give the joints/ligaments in my wrists a chance to recover. The possible alternatives outlined were not great (significantly advancing the age at which I will get arthritis), so I did as I was told. This obviously had a major impact on my ability to work on this, as real life work took priority, so things have not advanced as much as I (or anyone) would have liked. The good news is that I'm more or less back to normal now. Or sort-of: I have to wear some quite annoying wrist supports for computer use, but that's much better than the previous situation.

There are also additional circumstances which I can't really say anything about.

 

[lmaccount]

OK, so the thing that I can say: I've had some health issues - If people haven't noticed, I haven't really been active here over the last few months, and that wasn't by choice. Basically I suffer from hypermobility, and I was advised by my doctor to reduce my computer usage for a while to give the joints/ligaments in my wrists a chance to recover. The possible alternatives outlined were not great (significantly advancing the age at which I will get arthritis), so I did as I was told. This obviously had a major impact on my ability to work on this, as real life work took priority, so things have not advanced as much as I (or anyone) would have liked. The good news is that I'm more or less back to normal now. Or sort-of: I have to wear some quite annoying wrist supports for computer use, but that's much better than the previous situation.

There are also additional circumstances which I can't really say anything about.

Consider a style of keyboard that is inverse slanted instead of the usual style that bends our wrists unnaturally.

 

[habisain]

Consider a style of keyboard that is inverse slanted instead of the usual style that bends our wrists unnaturally.

While the sentiment is nice, the problems I have are a bit more complicated than a single wrist bend. I'm not too bad on flexion and extension movements, but supination, pronation and ulnar/radial deviation are all problematic for me. Some of the different keyboard layouts can increase these bends, which would be bad for me. Instead, I benefit from things like 10-keyless keyboards and whacking mouse DPI to 10,000 (where possible, obviously).

 

[rizo]

While the sentiment is nice, the problems I have are a bit more complicated than a single wrist bend. I'm not too bad on flexion and extension movements, but supination, pronation and ulnar/radial deviation are all problematic for me. Some of the different keyboard layouts can increase these bends, which would be bad for me. Instead, I benefit from things like 10-keyless keyboards and whacking mouse DPI to 10,000 (where possible, obviously).

Glad to hear you slowly back to normal- kinda.
Take your time as they say: Better late than never.
Also, this has to be the first time I've actually heard of a valid use for the crazy high dpi on mice so thanks for that little chuckle, since I imagined the mousoe boxes "10,000 dpi for when your joints are fucked" If it works, it works. Hope the other reasons are sorted sooner or later. Best of luck.