Feature suggestions and discussion [READ THE FIRST POST]

@Clover the Clever  
Ah, gotcha.

Maybe this is a thing already, though I suspect not, but I see a lot of instances where two people upload the exact same picture within minutes of each other.
 
Why not implement a system that hashes the raw uploaded file and when that matches along with the automated perception matching, automatically reject the upload as a duplicate?
 
That might cut down on some admin work.

@Jondor  
They already have that. It recently prevented me from uploading a third copy of the picture I imagine you’re thinking of.

@Jondor  
Problem is we need the images to actually be exact. Tumblr uses a 50 server CDN and each serve a different file even if it’s the image is the same (and often it’s not, different compression levels across servers); DA does the same, plus changes metadata intentionally every so often, etc.

And why not implement a neural network using the thumbnails to automatically predict who the ponies are in the image? Because science.

@Background Pony #4B60  
Probably because devs have better things to do than spend time making that, or server resources for it. Especially considering the most common characters have 2-3 characters aliases.
 
uploading with "m6,ab,luna,tx,sus,eqg," gets you tags for "mane six, twilight sparkle, rarity, applejack, pinkie pie, fluttershy, rainbow dash, apple bloom, princess luna, trixie, sunset shimmer, equestria girls", which IMO is neater than accepting taggers/uploaders are completely lazy and hoping a program recognizes a yellow and purple smudge as Fluttershy consistently; we get lulzy enough results from the perceptual dedupe as is.

@The Smiling Pony  
Fair enough. You can do a lot of cool stuff with this kind of “big data”.

@Background Pony #4B60  
Man, Derpibooru is not Google.  
And even Google has issues doing that.

@The Smiling Pony
 
That would explain a lot. Nevermind, then.

@BigMax  
It’s actually pretty simple to do and I did do a prototype some time ago, it’s just a matter of hooking it into Derpibooru, which is more faff and extra overhead the codebase really doesn’t need.
 
Several of the Derpi devs have experience in this field, or the periphery, myself included. It just comes down to time and stress - we do Derpi for fun (ahahahahha).

@Clover the Clever  
Now that’s interesting - what would the mis-recognition level be?

@Clover the Clever  
Imagine doing something like this with ponies

how about the ability for a user to un-anonymize their own comments? I know I’m not the only user who accidentally posted anonymously once or twice

Can you make it so that pressing S to go to the source opens it in a new tab like it does when you click the link?

It would be cool if we could fave and upvote using keyboard. (Not downvote, because it’s good that you have to put some effort into downvoting.)
 
And while we’re at keyboard navigation, I was wondering why is the prev “j” and the next “k”? In Vim k is up and j is down. This doesn’t bother me, just asking.

@Background Pony #BC7C  
Once, you had to use the arrow keys, but that interfered with scrolling (I think).  
Anyway, Derpibooru is a web image board, while Vim is a text editor (and frankly I don’t get why two horizontally adjacent keys should mean up and down instead of left and right), so you can’t really compare their behavior.

@BigMax  
Nobody understands why vim uses those keys.
 
Seriously, first thing my vimrc does is make arrow keys behave normally.
 
(Second thing it does is open nano ¬_¬)

@Clover the Clever  
That’s why.
 
Fortunately usually arrow keys are enabled by default. Depends on distro.

@Background Pony #BC7C  
You must be kidding me… To follow the standard of an old ass Xerox machine from 1977?!

@BigMax  
Yes, that’s why. Everything must be compatible with the 70’s tech.
 
I always love (no, not really) using the “old-school” Unix/Linux text editors. They’re all either so cryptic it takes eons to learn even the basic functionality, or they have weird keybinds that cannot be used properly if you aren’t using the US keyboard layout. Fun fun fun.

@JP  

 
Jokes on you, I can’t use Atom, because it has issues with my keyboard layout. (“…for the 21st Century”, lol)

@Background Pony #BC7C  
Atom is still immature, but quite promising indeed (I can’t use my user defined keybinds that are fundamental for me to type ` and ~ without using Alt sequences - dammit!).  
Brackets is doing better (version 1.1 is out), but I still have problems with it (where the heck did all the icons go?!).
 
Edit: it looks like installing Inkscape made Brackets lose its icons - WTF?!

Under the “Plan on doing section,” I see that you guys have “Improve reverse search engine” listed. Does this include a reverse image search API?
 
If not, I’d like to formally suggest adding an API for accessing the reverse search engine. That kind of API would be really useful for programmatically tracking down the source for rehosted images, and would provide a much needed service for the community.

@OldTalesChangeStyle  
That’s quite an insightful request.  
Are you developing an application that would make use of that?

@BigMax
 
I’m currently working on an open source and easily portable Reddit bot for /r/sunsetshimmer. Among other features, I had planned to use reverse image search to help automagically find and post the sources for images hosted on common rehosting sites like imgur, so that artists get the proper credit they deserve and users can easily find more of their work.
 
Unfortunately, reverse image search seems to be something that’s fairly closed off from a development standpoint. Both Google and Bing have well developed reverse image search features that are quite accurate, but neither of them has a public API for accessing it. It’d be possible for me to scrape from Bing’s results, but that’s fairly unethical and disallowed by their terms of service (although Bing has illicitly scraped results from Google before, so perhaps it’d be apropos). The only proper API I could find was TinEye, but they charge quite a bit of money per query and only seem to give mediocre results. As such, until I realized that Derpibooru had their own, in-house reverse image search, I had given up hope of realizing that particular feature.
 
If Derpibooru were to implement an API for their reverse image search, I could fairly easily implement it into my in-progress bot along with reasonable rate limiters if needed, and I imagine that other developers (including those who created similar helper bots for other subreddits) could make good use of it as well.