Software Sets New Record for Solving Jigsaw Puzzle

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
Is it just me or does solving a jigsaw puzzle sound like it should be easy for a computer to do? Well, apparently it isn’t easy and “setting new records” still means the puzzle looks like ass when the computer is “done” with it. :p

Completing jigsaw puzzles is a challenging and popular hobby, but now scientists from the Massachusetts Institute of Technology (MIT) in the U.S. and Tel Aviv University in Israel have for the first time developed a probabilistic graphical model to solve a jigsaw puzzle consisting of 400 square pieces.
 
WTF, how is that even considered a monumental, accomplishment? And, that image reconstruction is a joke... :|
 
finding the shortest tour through all nodes of a graph should be an easy task for a computer too, but apparently it isn't.
 
The real problem here is the ability for the computer to detect what image fits where.
If they get that right, all they need to do is atribute a x/y coordinate to each piece, and then use a numeric algorithm to solve the puzzle.

But most likely, it fails to diferenciate between similar images.
 
didnt click the link but I dont see how solving jigsaw puzzles would be trivial
 
The only way I would consider those puzzles successfully completed is if I hand my beer goggles firmly in place.
 
If you look at the example pictures in the article this isn't your typical "jigsaw puzzle" this is more of an image slices in to tiles, 400 I think and the computer told to reassemble them.

They don't have the usual sides people usually use them and don't forget computers can see. The program has no concept of what the image looks like. I guess a test would be to take a picture and cut it into 400 peices and feed it into an auto-sticher.
 
I think the images they posted were the initial rough image that the program sketches after it sorts the puzzle pieces not the finished product. It then uses that image as a color pattern guide to determine the most likely place a puzzle piece will go. The article details this, but there was nothing clarifying this on the images.

If a puzzle was just about shape recognition and putting it together, it would be a fairly trivial task for a computer. However the computer doesn't know what the final picture will look like and it has to match the most likely color combinations together.
 
Anyone notice it reconstruced the building on the wrong side of the road? LOL
 
Their puzzle pieces are SQUARE! That'd be a bitch with the clouds ...
 
They used the same tile multiple times in the sky needle picture. Seems like eliminating the possibility of using the same image multiple times would increase the accuracy of the final image.
 
The same computer used the school logo for a puzzle. SUCCESS!

MIT_lol_puzzle_paint.gif
 
isn't there software that can reconstruct text from shredded bits of paper? after that site it all sounds fake, like government powers on TV.
 
Perhaps all the puzzles I've seen were on easy-mode, but there was usually an image of what the final image should look like on the front of the box.

If you gave the box to the computer, it would probably have your solution within milliseconds.
 
Perhaps all the puzzles I've seen were on easy-mode, but there was usually an image of what the final image should look like on the front of the box.

If you gave the box to the computer, it would probably have your solution within milliseconds.

If the computer could see what the final image was like, it would be trivial because the computer would just match up all the tiles with the correct ones in the original image. That code could literally be written in less than 20 lines.

The problem that many people on here fail to grasp is that the computer has no concept of what the final image should look like. This would be bad enough for a human, and we can generally recognize things like buildings and clouds. But a computer can't. Hence the problem. Considering that, what the guys at M.I.T. did is pretty astounding.
 
Back
Top