An Approximate Differentiable Renderer

Опубликовано 27 июня 2016, 20:53

Although computer vision can be posed as an inverse rendering problem, most renderers are not tailored to this task. Our framework makes it simple to express a forward graphics model, automatically obtain derivatives with respect to the model parameters, and optimize over them. Built on a new autodifferentiation package and OpenGL, OpenDR provides color, depth and boundary renderers, and promotes the ability to construct functions of the renderers. We demonstrate the power and simplicity of programming with OpenDR by using it to prototype a system for estimating human body shape from Kinect depth and RGB data.