Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[hops]

So the answer is that xkcd's style will become oil painting.

[Frumple]

60% certain oils would actually be easier for a computer. Far less cohesion needed to make it look good, from what I recall. You can be hella' sloppier and still get something fairly nice. Not something I've messed with much, by I've got relatives that actually do painting workshop type things and they'll occasionally laugh about how little it takes to get paid to do basically nothing skillful

Alternately, just take the doodle thing (or similar software) linked and apply one (or several) of the oil brush(es) already commonly available to image editing software :V

[Starver]

Not something I've messed with much, by I've got relatives that actually do painting workshop type things and they'll occasionally laugh about how little it takes to get paid to do basically nothing skillful

I've dabbled in painting (acrylic paint, both watered down as backwash and then shapes and details slopped in as with 'oils'), but I find it hard to subvert my analytical self-criticism by bold artistic strokes, so they just up looking awkward.  (Not naive, impressionist or cubist, and naturally I have not the skill to make them classically 'photorealistic'.)

Alternately, just take the doodle thing (or similar software) linked and apply one (or several) of the oil brush(es) already commonly available to image editing software :V

Which is what makes me sceptical about the "takes several hours without the GPU" claim.  GIMP does well enough, with various styles, and although that's "photo->art that looks like the photo" rather than "sketch->art that looks like it could be reduced to the sketch", I think I can spot some of the key decision-making processes in the software (colour contrasts/blending; image parsing to recognise tree-forms, rock-forms, sea-horizon forms; feature addition (e.g. distant boats) as apt to the prior understanding) and don't doubt that someone could use a Scheme-based filter to do something similar.  (Yes, most of the processing time taken will be from the 'image parsing' part for contexts, I know...)

Also, forget oil brushes... try tapping the toothpaste pipe.

[PTTG??]

Ugh this neural doodle thing is so annoying to set up. How is "python3-dev" different from regular python, and why is there no way to install it on windows?

[Bauglir]

How is "python3-dev" different from regular python, and why is there no way to install it on windows?

Well, apparently it's got modules for extending the Python interpreter and, more likely relevant to what you're doing, embedding Python in applications. Also, it's definitely Python 3, while what you're used to as "regular python" might be Python 2. Why is there no way to install it on Windows? I don't know. I imagine it's part of something else that works just fine on Windows or something, but maybe there's just not enough people who want it and are unwilling to work in Linux.

[Arx]

I've noticed a definite trend that neural net stuff assumes you're a Linux user. Python3-dev is really easy to install on Linux, but I don't know if it even exists for Windows.

[Telgin]

Yeah, Linux is more or less entrenched in the academic world in computer science.  A holdover from the very early days I suppose, even if increasing numbers of grad students prefer Windows these days.  Even my advisor has told me that he's considered replacing Linux everywhere in his lab because he's absolutely sick of being a system admin and fixing stupid things when they break.

But then you get problems like not being able to install packages you need because academic software is almost exclusively written for Linux.

I'd guess that the python3-dev package is the headers and libraries needed to link the Python 3 interpreter into a new program.  That means that it probably doesn't really exist for Windows.  It's rare for anyone to provide compiled libraries like that for Windows, although at least many people are kind enough to provide fully compiled binaries for it.  So you can get the Python 3 interpreted pretty easily, but if you want the libraries like that you're probably going to have to download the Python 3 interpreter source code and compile them yourself.

Kind of interesting though.  Are you having to compile something already in Windows that needs the package?  If there's a Windows version of whatever you're working with then I'd expect someone to have provided either the libraries for Windows or directions on how to compile them.

[wierd]

There are windows python builds, yes.

[Starver]

Yeah, Linux is more or less entrenched in the academic world in computer science.  A holdover from the very early days I suppose, even if increasing numbers of grad students prefer Windows these days.  Even my advisor has told me that he's considered replacing Linux everywhere in his lab because he's absolutely sick of being a system admin and fixing stupid things when they break.

It's been a long time since Windows 'lacked' a system admin (ignoring the very modern possibility of "running as root/administrator" as a risk that needs to be carefully considered before taking it), and perhaps also arguably had correspondingly more "stupid things that break", certainly much more so than a well pre-configured *nix system.

Spoiler: Back in my day (click to show/hide)

...Linux was something we surreptitiously put onto Windows(pre-95)/DOS(<=6.x) lab machines to get the same look'n'feel as the X Windows System terminals, rather than just using the TTY emulators to get just the non-graphical login to the Unix servers we had ready access to.  It was one of the few things we could do (until they locked down the BIOS) to fully remodel the computer, which was normally equipped with a handy utility that compared and corrected all important parts of the directory structure with a template discovered over the IPX/SPX network.  You could delete the C:\WINDOWS or even C:\DOS directory and reboot the machine back to 'standard'.  Formatting the disk or messing with root system files was sometimes a problem (and definitely the hidden .SYS files) and of course reinstalling the partition with Linux went further than that...  And once the BIOS got regularly passworded (and the cases already physically locked) you started getting obviously-broken computers that a 'white hat' student couldn't even get back to normal again, as they could previously.)

It was only as the NT branch made its way (via 2K->XP) into the 'regular' PC market that there started being any point to user access control (although the actual User Access Control in Vista was a clear overkill with an annoying and badly implemented mode of working!).  And physical access to a PC (and, even with UEFI, boot-protection can be bypassed, as I have had to do it myself several times recently) lets you get around most 'admin-protected' systems, including Linux (even if you have to rely on JohnTheRipper to give you a toe-hold into at least one of the useful system passwords recoverable via a privileged Run-Level), and exceptions to this rule (see the current story about the 'terrorist iPhone' and the FBI) tend to be in bespoke hardware, only...

[wierd]

Windows level secure passwords are a joke. Especially if you can boot the system on removable media.

Offline NT password recovery.

It's a tool based on a minimalist linux. Fits easily on a USB stick, CDROM, hell-- even a floppy diskette, if you have a system that can still take one.

It lets you put an arbitrary user into the Administrators group context. Password or no password. Also lets you reset any arbitrary account password to blank.

GREAT for regaining control of the home PC from punk kids.

Also terrible for network admins trying to lock down a computer. Whole drive encryption is about the only real way to defeat it.

[Telgin]

Yep, a well configured Linux system is probably more secure than an equally well configured Windows system.  And you do have to still set up users and such like a system admin.

My advisor's problem is mostly that in order to use Linux even in a desktop setting you have to have an almost system admin level of knowledge.  His words, not mine.  I can't really argue though, at least with the systems we have.  That could just as easily be a problem with things outside of his control though.

My favorite Linux problem was when the Ubuntu login screen would freeze.  Or just do nothing when I tried to login.  The problem was that my user directory was exceeding the disk quota.  I only discovered that by chance after deleting Chromium's gigantic cache and trying again.  Or the fact that it would freeze when the NFS mounts failed to mount at boot time.  That required going into the recovery console, remounting the file system in R/W mode, modifying the /etc/fstab file and several other things just to get back in.

I'm not saying Windows doesn't have problems, but... most grad students, even CS grad students, aren't really equipped to deal with stupid stuff like that breaking.

But wow I feel like we've gotten off topic here...

[wierd]

Like basically any modern system, Linux can be disk imaged, and even network booted over pxe.

Basically, the PXE bios in the NIC executes something like memdisk, which tftps a small initrd into the workstation, then boots it. The initrd loads nic drivers and the nfs kernel module, then mounts all the remaining filesystem as NFS over the netowrk.

Boom. Network managed linux on a diskless workstation.

If the mounted remote is something like a squashfs filesystem, and RW is accomplished with -bind with tmpfs, then all changes made are volatile, and will be lost upon reboot.

Linux can be made VERY hard to break.

[Bauglir]

-snip-

I was gonna post about how NTLM passwords are much better than they used to be, to the point that they can't really be brute forced, but it looks like this just sorta goes "Yeah, that sounds like a lot of work, I'll just use shenanigans to make the password whatever I want it to be."

Welp.

[Starver]

I was gonna post about how NTLM passwords are much better than they used to be, to the point that they can't really be brute forced, but it looks like this just sorta goes "Yeah, that sounds like a lot of work, I'll just use shenanigans to make the password whatever I want it to be."

Have they sorted the old problem (over a decade old, so probably) whereby longer passwords were split and hashed separately, so you only need to brute force half-n-half?  (Rather than, you know, at least using the result of the low-order hash to salt the hashing of the higher-order one...  And there's better ways of doing that, even!)

Not that it matters to me, I've built up quite a number of useful tools for Windows, including the one that just goes straight into the Hives and nukes/resets passwords, unlocks locked accounts, adds accounts to admin groups, etc...  I no longer (much to my regret) really have to know how they do this, once they've proved themselves as worthwhile.  Very few of them enable subtle "Spooks"-style cracking, but when I'm trying to get around malicious damage/locking and am doing it on behalf of the legitimate user that's not a problem.

(I have rarely had to crack through Linux security - for reasons ranging from their being less represented amongst the kind of person who needs such simple problems solving through to them being 'higher hanging fruit' both technologically and again w.r.t. to their liveware vulnerabilities - and I'd probably have to gather the tools I used together again.  The last copy of John I had handy got 'quarantined' by an overzealous AV, because it was a "hacker tool".  Which it is, of course, give or take.)

[Sergarr]

AlphaGo has won a third game against Lee. That means it has won the overall match, and the title of world Go champion (and 1 million dollars in prize money) goes to AlphaGo, after they play out the remaining two games.

An interesting commentaries on the play of the machine:

It would appear that Alphago gets noticeably better between each match. Imagine if we had expert systems like that.

An interesting thing about these matches is that Lee Se Dol and and Alphago are all roughly evenly matched/hard to tell throughout the early/mid game or exchange who's in the lead; until suddenly out of the blue Alphago get a crushing gain in territory in the end game.