Using Neural Networks with 3D Data One of the papers that inspired my current work with using AI to assist further in character rigging was "Fast and Deep Deformation Approximations" by Bailey et al. Using a blog post on 3DeepLearner.com, I was able to recreate the project described in this paper. I feel this was a necessary step in designing my thesis project because it deals with 3D data and how to get this data through a neural network. The data acquisition and pre-processing is likely to be one of the hardest parts of the project to design, so all the information I can find on how others have done it should be helpful. The gist of what is happening here is that the authors wanted to use deep learning to replicate the complex deformations of a professional studio rig whilst making the deformation more efficient. This was done by creating a neural network to evaluate the possible deformations for each joint and its range of motion. Then the network's model, o
Boom. Displacement. I didn't get to spend as much time on this as I would have liked to this week, but I did have enough time to add at least one displacement texture to the rig that triggers with one of my driven key expression controls. What I was able to do was export the base mesh of my character with her brows lowered into the anger pose. I took that into ZBrush and sculpted some little wrinkles onto the eye brows. After exporting the maps and setting up a shader in Renderman for Maya I could insert a Multiply/Divide Node into the shader graph which would just multiply my displacement map by a value from 0 to 1. This essentially just lets me turn the displacement on by multiplying by 1 and off by multiplying by 0. I'm no master sculptor, but for this test, this map did the trick. This is a pretty simple setup since I'm only using one triggered displacement. However, I realized that if I had a base-displacement for skin pores or anything like th