Anyhoo, here's our first in the series :
I made images like this back when I was working on Progressive Meshes. It was hard to find good quality high poly meshes to test with back then, so I had the idea of running the PM on an image as a mesh. I load a bitmap and make two triangles for each pixel, and put the pixel colors on the vertices. Then I just run the normal PM collapser in Galaxy and it automatically removes flat areas.
You can see the PM does a pretty nice job of edge detection and whatnot; compare to the original :
It was a great test of the tech, but I thought the images were really pretty too. And amusingly they get more interesting looking as the triangle count gets lower and created the cool cubist effects of big jagged triangles of color gradients. (in the above picture it's been taken from 128k triangles down to 1768).
Here's a bonus, simplified Lena in wireframe :
5 comments:
Ooh, I'd like to see that in high res.
I'd like to see you do the PM on a movie. You would probably have to chose vertices based on motion as well as color.
Also, it would be interesting to take the diff of the PM image and the original, and store the diff as a compressed texture and see how good the image looks. Maybe the mesh of the image is an "envelope" that captures both low/high color values (or equivalently, mean/spread) and the compressed "diff" texture interpolates in that range.
That chairlift video is amazingly sweet. Thanks for linking it (and thanks Won)!
My pleasure. It is easy to find interesting things when you have an in at YouTube. Still working on the "hump matrix."
what happened to these images?
Post a Comment