Learning materials for training on 3d rendered images, parameters, and geometry

I am a 3d technical artist wanting to create tools in 3d applications like houdini. I am interested in things like generating lights by evaluating image lighting, and deforming geometry based on skeleton joint rotations. I have taken the specialization for linear regression, advance learning algoriths, and unsupervised learning. Any recomendations on learning videos or courses i should go onto.

This video explains more my interst.

Most of your audience here is not going to know what Usd and Pdg are.

It appears you’ve already created the very system you are asking for recommendations about how to achieve.

This leaves me confused.

Thanks for you reply tmosh. I have partially created it. I feel there is a gap in my knowledge where I do not know how to design neural networks properly. For instance, I can get 3 color floats to work good, but when I also try to train 3 color floats & 4X4 transform matrix floats (this would be the transform matrix of the light), for a total of 19 floats, my results are not so great.

Since my post I have found “Neural Networks And Deep Learning” course that I feel will help fill my knowledge gap, and am going through that. If you have other recommendations I am happy to hear them.

What I want to get good at is using deep learning to create models that people can use as tools in their process of creating 3D animation and vfx. Examples are cloth simulation, gas sim, lighting setups, and rigging deformations.

I am old hat at using 3d software and scripting, but a novice in machine learning.

The DLAI “Deep Learning Specialization” would be useful for you to attend.

1 Like