so i just had an idea for a test it would sort of be the calgary corpus of image recognition think of a 20 sided die, where every side has the number it represents printed in english (or handwritten, or a different language, the dice can be modified but the principle is the same) this die would be in a transparent box with a camera above it looking down the box would be shaken, rolling the die the camera would then take a picture
the goal for any algorithm attempting this problem would be to determine the number that was rolled
anyways i have no experience with image detection at all, how would i go about finding people to try this problem
but stackoverflow is strict on the "questions and answers only" format, ill probably get downbumped if i start a challenge thread
Dominic Ortiz
seems like the box is an unnecessary complication, the images could be pre-generated, then your looking at general OCR. You don't need people to try to solve this problem, you just need various algorithms, free and proprietary software to try to solve the problem, then compare the performance.
Owen Martinez
seems like the die is an unnecessary complication, the rolls could be pre-generated, then your looking at general OCD. You don't need to try to solve this problem.
Hudson Martin
well the whole point of the test really would be to extrapolate and identify text from a real life image
we could also have it so the camera is not neccessarily looking straight down at the die, just from an angle where a human would also be able to identify the number rolled
the main idea here is that the difficulty of the problems stems from being able to make out which number is on top (rolled) and being able to extrapolate and interpret text data
so only half of it is really ocr
Aiden Cox
There's also codegolf.stackexchange.com, perhaps that's a good place to post it?
Jack Young
still doesnt look like the right place to post it, but this looks pretty cool for casual code challenges thanks user
This was already done as a student project a decade ago.
Isaiah Thompson
are you shitting me
Ryan Miller
There's scikit for python, or if you want to do it by hand you can too, although with a 20-sided die it'll be nowhere as easy as with one with the dots. Anyway, choose some image recognition lib and try it, make sure the dice can be seen clearly by the camera.
Correct.
Blake Sanchez
i was thinking openCV
Jaxon Garcia
the serious problem here is to isolate the text from the top, because 9 other sides are visible at any given time
Josiah Smith
But you know the text beforehand. The one on top should be the best match since the others are at an angle. You could also easily cut the top since you know the size of the die.
Jason Reyes
what about from an angle or a distance
Jaxon Baker
The distance is also a non-problem since it's known (even if the die isn't dead-center in the box the distance to the camera is pretty much the same). As for angles I don't know, I only used ARToolkit I think it was and it uses/used some patterns so they were easy to identify, worst case you could do the retard solution of rotating it and picking the best.
Colton Nguyen
Was it the one with a bunch of dice being rolled at once?
Eli Butler
Character recognition. It has been done many times before.
Isaiah Cruz
What is this, 1970? This is a trivial problem.
In fact you don't even need to do ocr, since there's only 20 possible sides. Just take a photo of each and match to them. Or even more brute force way: For each side, take 100 photos of die from different angles, for a total of 2000 photos, then just find the most similar one among those with minimal processing. And what's this dice camera faggotry? Just generate it with a script. It's just a shitty captcha anyhow.
Noah Gutierrez
Probably PRNG. Still, OP is a fag
Aaron Diaz
Paint sides in different colors, detect color in horizontal plane. Are you trying to reinvent low-entropy PRNG?
Tyler Smith
well im getting the sense that this has already been though of and solved lets shift gears and just talk about image recognition in general
