This is an alpha version, I hope to improve it over the next week by incorporating ConvnetJS, to label pixels in mixed superpixels (Superpixels with both positive and negative labels). And incorporate additional functionality to such as blurring the boundary.<p>The current restriction on the size of canvas 640 X 640 (Instagram size) , will also be removed over next few updates.
It was hard to get an idea of the final result in the demo gif because the last frame disappeared so quickly. Here it is <a href="http://imgur.com/fduYXCi" rel="nofollow">http://imgur.com/fduYXCi</a>
Was working on a similar problem, to fetch color of cloths from images(including models and background) in an e-commerce website. This implementation gives the best result so far.
In the readme file, you have "License: Copyright 2015, Cornell University. All Rights Reserved." Does that mean there isn't a public license to use it?
Great demo. Out of curiosity, why not compiling the C algorithm with emcc into asm.js and probably get better performances (but a much larger js file) ?