Recently at Siggraph 2010, Johannes Saam and Chris Cooper presented an overview of deep image compositing at the Foundry user group meeting All the videos show the concept working inside a prototype written at DRD Studios.
1. The basics:
This video contains the basics of deep image data. You see an alpha image containing deep samples. The meta data in a deep image can be used to generate a point cloud of samples in world – space. Flattened images can be analyzed by drawing a graph that shows the sample distribution over x and the values over y.
2. Volumetric compositing:
Here you see the basic usage of a deep image. A solid object with transparent edges is inserted into a volumetric rendering. Rather then using the traditional merging operations, images can be integrated into each other. The depth and optical density of the samples can also be manipulated at this stage. Holdout mattes are generated on the fly and then used to comp the two images.
3. Deep data generated in nuke
Deep data can be generated everywhere along the pipeline. In this example we use the depth output of a nuke scanline renderer to generate a deep image. This image can still be freely manipulated and combined with the pre-rendered image.
4. Merging Deep data
Deep data sets can be combined into larger sets. This can be thought of as an additive process. Samples are added to each pixel and a more complex scene is created.
5. Volumetric noise
3-D noise can be applied to the image domain which generates a consistent field of noise to generate atmospheric effects. A grade node is also introduced. The amount of grading can be regulated with a curve over the depth away from camera. This is just a prototype for much more complex operations. Instead of constraining effects to flat masks we can generate 3D masks with ellipsoids, etc.
More videos are coming to show the tools in shot contexts and present new, more flexible work-flows!