Could you pls, Jow Forumsods of tech, help /v/ on figuring out what the fuck is cdpr showing us here

Did you get some OCR fuckups and can't see it whole?

Haven't started OCRing yet, first I'm filtering every frame to get every line of text.
Also I might have to train tesseract on that particular font, I think it's Terminus?

Its obviously for their Cyberpunk game

bitcoin address to pre-order the game

already seen the image generated with some distortion.

its a photo of co-op multiplayer in their new game. has a ui similar to world of Warcraft party windows etc.

could you sauce us the image, man?

Can't get good enough OCR to work, I have to train tesseract and that's a pain. Or find an OCR engine which isn't a pain to train on a certain font. Or roll my slow-ass own.

maybe this helps
imgur.com/mJq2ItH
I am reading that people believe it is looping

No I already have enough frames. Not sure if anyone can tell it's looping without anyone having done accurate OCR on it. It's probably a sequence of PNGs though.

What did CDPR mean by this?

Attached: cyber.png (1260x1020, 725K)

they are 5 letter codes to gog.com but they are taken instantly or it's bugged

Alright I'm rolling my own OCR.
If someone would like to help me as I'm writing it, could you manually type all the characters in this screenshot (with line breaks) and post them to pastebin?

Attached: 1520617191436.jpg (1920x1080, 630K)

Or as many as you can actually, I don't really need the whole thing but more is better.

errbody wanna know the secret, but aint nobody wanna do no boring ass data entry my dude

here's just enough so i don't feel guilty, but not enough to be useful
S8WaKBi3BZW4sGTHotm9vbjKZPMTOTlgmGyHGXhT6aI6LR6ToFWOFuA7jmBxoDyD5gzEGZRn

Imagemagick + gocr and fix errors

$ tesseract mJq2ItH.jpg mJq2ItH -l eng
Tesseract Open Source OCR Engine v3.05.00 with Leptonica
$ < mJq2ItH.txt | ix
ix.io/1lm6

obviously tesseract's default data files weren't trained for this

>drup_package
>mixing B and Ds

now what
GNU coreutils online help:
Full documentation at:
or available locally via: info '(coreutils) base64 invocation'
karl@karl-iv-hrr:~$ base64 -d cyber.txt > out.txt
base64: invalid input
karl@karl-iv-hrr:~$ base64 -d cyber.txt > out.txt
base64: invalid input
karl@karl-iv-hrr:~$ base64 -d cyber.txt > out.txt
base64: invalid input
karl@karl-iv-hrr:~$ base64 -d cyber.txt
�PCF�J
�q�IHDR��Gbase64: invalid input
karl@karl-iv-hrr:~$

typed 1-4 lines
iVBORw0KGgoAAAANSUhEUgAAB4AAAAQ4CAIAAABnsVYUAAjxYklEQVR4AYzYBXPbyrcA8JQS
s0VmjplR5khmhpgxdhx0kzTc9DK0vfjud35HVv6+KTyY+c300eosaAUzq+ce/xD95zhEQCX
BwMOtwDYXUKHW+RyIsDjxoDxiy6Jg+HsJaHQuD2CV1eaMgDy+Z8m5th9wiWeOAxdiM2l/hL
bD6wunhPWZzcL0E9m/zFKFav/JLdI3J6sc/Axbq90Htdfi8W80EBHxoK4I0C/EhE9CgmZsUS

How about we roll to type line?

oh, boi

Attached: file.png (640x372, 389K)

i.imgur.com/fWqwARd.png

fug, sorry

Attached: file.png (662x608, 28K)

based red pilled poles

OCR was kind of unreliable in my tests, especially with the transient frames (where the text changes). I achieved something much closer to accurate using perceptive imagehashes and just cropping the characters (it's monospace). Code is here:

gist.github.com/cryzed/8874f9cc2a4ccf7b10ca78bc115982cc
www41.zippyshare.com/v/3rF0BJWq/file.html

The idea is to simply use ffmpeg to extract all frames from the file, analyze every frame and stitch together the parsed base64.

Once I find a non-shitty imagehost where I don't have to register first to upload something, I'll upload the characters-folders.

Current problems are stitching together the text properly with the next frame and making sure that the characters are recognized properly at all times (even during a frame-switch).

Nevermind, the images.zip is already linked (hosted on zippyshare). I don't quite feel like finishing this, since it just feels like a waste of time now. Maybe someone else can pick up where I left off.

the whole thing is here
i.imgur.com/mJq2ItH.jpg

Bump

>watching this gibberish for 3 more hours

cybershit2077 is gonna be a shit game. What are CD peoject red hiding by feeding us this for 10 hours? garbage company

pastebin done by ocr (lots of errors) 0DpWJ8Bj

i.miosblog.de/cdpr/
hi res screen grabs

I made a base64 conversion on a screenshot I have.
It was 60-70 000 lines with a slightly wider lines then they use
they print them at a rate of 20 per minute
it will take 58 hours to dump the entire thing
what do we use to decode it? ffmpeg and some image slicing?

are you sure user?

the guy that posted this said the text loops after a while

Hi everyone, I just got here, but here are my OCR results pastebin.com/dtsxHfkE

I tried decoding from Base64 and saving as PNG but the decoder gave me shit about it having non-base64 characters.

>cybershit2077 is gonna be a shit game. What are CD peoject red hiding by feeding us this for 10 hours? garbage company
probably shifting again to some shilling graphics library in the last moment, like barellyworks on witcher 3, and enjoy their money while the idiots are stuck in super-market graphics.

Shit, nevermind I fucked it up. Trying some more.

we'll be waiting, user

Alright, now we need people to verify if these images are really looping and if the characters are correct

Thesw anons have made some good progress already, but we will need more if we actually wanna decipher this

OP, can you recruit more /v/irgins to help us out?

I'll be done with my OCR in 10 minutes approximately, then I'll just need to dump the video frames and do some minor stuff in order to automate reading everything.

Split the capture into ABC grid and commit sections to a repo?

>>/v/429329179

Yay I have a 100% character recognition score on my test.
Time to dump these frames!

Thanks mate

So you're good, no more transcription needed?

try first lines and compare to this

I HAVE A PNG
But my image viewer is complaining about a premature end of file, but at least base64 seems to think it's a valid base64 stream!

can you pastebin encoded txt?

Not until I know what it is before anyone else, sorry.

neat
open it irfanview
it deals with broken pictures very well, lad

Is OCR really needed for a fixed-width font like this?

>looping
as far as I can tell, the stream appears to still be outputting new data

I'll try but I just noticed a little bug in my code as well. At least mediainfo can read the header fine.

OCR is just character recognition. I ended up implementing a super primitive one which literally does a pixel comparison with and "example" font. Its only problem is making the distinction between I and l in some rare cases and I'm not sure how to fix that. But I assume the png format is resilient enough to have some corrupted bits and still be somewhat readable.

l(low case) and i(upper case), i had top upper case more pixels,made special case test

its already uploadig something, you egomaniac.
You too late nigguh

Shieet)

>inb4 base64 encode of entire game folder

Where are you getting the source images from btw?
I noticed the images on i.miosblog.de/cdpr/ are missing a ton of lines, eg between
i.miosblog.de/cdpr/no_compress/0052.png
and
i.miosblog.de/cdpr/no_compress/0053.png
There's like 3 lines missing that you can see at 00:52:03: twitch.tv/cdprojektred/video/302423092

Using an OCR someone made from the images on there just gives this fucked result, so I guess there's probably more parts missing from those images.

Attached: not original file since 4chan wont accept it.png (1920x1080, 113K)

the base64 data stream ended just now

>an hour worth of frames
>only 56Kb
The fuck am I doing wrong now?

upgrade from 56k modem

That sounds about right, the PNG headers say it's only around 600Kb, and the stream lasted like 8 hours or some shit just to send it all

Yeah I was about to say that, also it was 97Kb, not 56.
I think my stuff works, I just need to dump more frames from the video which is taking very long.

it looks like shit kek

>stared at text for hours for this
fucking c l o s e d

no way

deus ex ripoff game

yawn

It was going at like 25-30 bytes per second, if that. Actually less because of base64's lower density than ascii.

(you)

Looked pretty cool

Okay I'm going to run my program run overnight to see if I get anything. If I don't I'll just give up.

You have a bug on line 86

Is the whole text actually assembled?
Why not first get all frames and cut out of the top line of text in each. Then compare each to filter out for similar duplicates. Afterwards OCR?

photos.google.com/share/AF1QipOg6ByRA_jkfgL8cmtKiF3L1tQ-oETifnt71Sc-xY80YKLUNJrFPXVXg-wzWDhFqQ?key=SWhVRXhMX1h0LWdmSkZxVmU4VFlQeFoxdjUybHFn

Yeah that's almost exactly what I'm doing.

the stream was like 9 hours of that shit text
after that this showed up, followed by gameplay of the new game
so all these giberish is probably a bunch of games keys or somehting like that

Attached: file.png (883x651, 436K)

It's base64 you brainlets

probably missing lines like every other collection people posted, you're best off grabbing frames and shit like said
going through frame by frame is a little inefficient though, instead you should figure out how many frames it takes for the line to change and go through it by that
i think it changes sometimes though, at least i noticed the stream pausing for a couple seconds sometimes.

also apparently CDPR devs red-teamed this and managed to figure it all out in 50 minutes, not sure how true that is though

If my calculations are correct the final file should be around 1Megabyte big.

Hate to break it to you, but nah, you're a fucking retard.

The base64 data is about 32 kB. The PNG is under 30 kB.

Is the stream available anywhere? I have a complete pipeline for extracting the frames and OCRing them, I just need the stream.

Yeah it's at twitch.tv/cdprojektred/video/302423092
You can use youtube-dl to rip it but it's around 24GB apparently, also last 48 mins or so is the gameplay trailer

Thanks, I'm on it. Will post progress and all data in here.

its going to be a picture of the devs dick and balls

Sweet, hope you can get somewhere with it user, those idiots on reddit/discord haven't got a clue right now (saw one of their main guys earlier talking about how he decodes the base64 using a site, copies it into notepad, then saves as .png.. jesus)

topkek

with the shit like "kill niggers" and "bush did 911" i seriously doubt it's anything legit and more like trolling

Not the other user, but the IDAT in the first "page" gives a length of 580 odd kB, which makes sense, considering the image is 1920x1080.

lel those are shoops, look at how every character except for the shopped ones fit neatly into columns

My biggest issue right now id downloading the damn video. If anyone have a good solution, please post it. Otherwise I have to dick around with these command line tools for a little bit.