I found two of these to be of direct relevance to real-time rendering:
Gaussian KD-Trees for High-Dimensional Filtering: This paper generalizes the approach used in the SIGGRAPH 2007 bilateral grid paper. Large-scale image filters are typically performed on downscaled buffers for better performance, but this cannot normally be done for bilateral filters (which are used in real-time rendering for things like filtering SSAO). The bilateral grid is a three-dimensional low-resolution grid, where the third dimension is the image intensity (it could also be depth or some other scalar quantity). However, bilateral grids cannot be used to accelerate bilateral filters based on higher-dimensional quantities like RGB color or depth + surface normal; this paper addresses that limitation.
Modeling Human Color Perception under Extended Luminance Levels: An understanding of human color perception is fundamental to computer graphics; many rendering processes are perceptual rather than physical (such as tone mapping and color correction), but even physical computations are affected by the properties of human vision (such as the range of visible wavelengths and the fact that human color perception is trichromatic, or three-dimensional). Most computer graphics people are familiar with color spaces such as CIE XYZ, but color appearance models such as CIECAM02 are less familiar. These are used to model the effects of adaptation, background, etc. on color perception. Current color appearance models are based on perceptual experiments performed under relatively low luminance values; this paper extends the experiments to high values, up to about 17,000 candelas per square meter (white paper in noon sunlight), and propose a new color appearance model based on their findings. I also found the background and related work sections illuminating for their succinct overview of the current state of color science.
Two more papers, although not directly relevant to real-time rendering, are interesting and thought-provoking:
Single Scattering in Refractive Media with Triangle Mesh Boundaries: This paper finds a rapid (although not quite real-time) solution to refraction in objects composed of faceted or smooth triangle meshes. The methods described here are interesting and look like they could inspire some real-time techniques, perhaps on the next generation of graphics hardware.
Fabricating Microgeometry for Custom Surface Reflectance: This one is not useful for rendering, but is just plain cool. Instead of using the microfacet model to predict the appearance of surfaces based on their structure, they turn the idea around and construct surfaces so that they have a desired appearance. One of the examples they show (inspired by Figure 1 in this paper), is a material with a teapot-shaped highlight! Well, with their current fabrication methods it is really a teapot-shaped reflection cast on a wall, but once manufacturers get their hands on this, all kinds of weird and wonderful materials will start showing up.
So far the yield of papers relevant to real-time rendering practitioners is disappointingly low; perhaps more relevant papers will show up when the official list is published. In any case, the early list of courses has a lot of relevant material, and I have reason to believe the final list will have even more good stuff on it. In addition, the Talks (formerly Sketches) program always has useful stuff, Will Wright is giving a keynote speech, and the Electronic Theater (which is back, renamed as the Evening Theater) now has real-time content, so there are more than enough reasons to attend SIGGRAPH this year (and it’s in New Orleans!). Registration has already started!