This is a basic list of ideas for projects, in pretty simple dot-point form. They're not in any order, certainly not order of being interesting. I'm leaning towards the "Infinite Precision" formats, as there's some really interesting things to try there, and lots of different possibilities, but I haven't found any evidence of people really trying it before... - Bézier curves on the GPU - Making object entirely out of geometry (no fragment processing, textures) - PROS: Easy to fit some sort of LOD in. Do it in geometry/tess shaders? - CONS: Probably not actually that useful in the real world. Unless doing lots of GPU stuff, probably already done and documented. - Implementing the triangulation part of Jim Blinn's GPU Gems 3 paper on the GPU - PROS: Well defined. - CONS: Probably no LOD. Rather boring. - Entirely in pixel shaders - Not really a good idea? It'd like be a really inefficient way of doing things? - Level of Detail for Bézier curves - This will either already have been done or be really difficult, I bet. - Lots and lots of maths. - "Infinite precision" by view reparenting. - PROS: Pretty unique and cool, as far as I can tell. Lots of different methods to compare, separate tasks. - CONS: Some nasty corner cases to handle. No-one's really done anything like this before. - Had a closer look at this yesterday: Postscript has a 32-bit float precision - IIRC, not quite IEEE float, trades a bit of range for a bit of precision, I think. - Thought a bit about implementation. There are some corner cases which would be quite interesting. - Quadtree sounds like it'd work, but would need some tweaks for "sibling" and "cousin" nodes. - Have to copy objects to the precision level we're working at (or one within a given precision range) - Lots of polygon clipping. If we do bézier curves it'd get close to impossible. :/ - People making postscript documents for (e.g) mandelbrot sets are complaining about the precision now. - Not really any research on high-precision document formats. This would probably be pretty uncharted waters. - Embedded shaders - PROS: Cool looking. Not too much work. - CONS: WebGL has done this already. Unless we try for some self-defining geometry shader thing. - Possibly try for a shader format which is guaranteed to terminate? - Embedding 3D models - Implement that U3D thing in libpoppler. Cool, but is it actually a project? - Precomputing opengl calls for a given document. - PROS: Performance, I guess. Possibility of lots of existing documents to convert? - CONS: Basically have to implement a whole document format. - Basically grab libpoppler or Cairo or one of the other open source PDF libs and make it basically dump raw GPU data for quick rendering. - Magnification filter for raster images. - Basically something that can scale photos while "generating" further detail rather than pixelating. - I smell fourier transforms or similar. - PROS: Some existing research. Interesting. - CONS: Probably quite hard. Not really any objective measure of success. - Need to check the literature in more detail.