D-CGA Shape

Differentiable Procedural Modelling

c++ python qt blender geometry processing procedural design machine learning

We introduce differentiable procedural modelling, a combination of inverse and interactive procedural methods. Using gradient-based numerical optimisation and automatic differentiation, we improved the usability of grammar-based procedural production systems. We build an application mainly focused on the designers: we allow direct modifications of the geometry and transparently adjust the grammar parameters to the modelling needs, while an intuitive UI provides feedback on the feasibility of the design. We built an application focused on designers, which allows direct modifications of the geometry and transparently adjusts the grammar parameters to the modelling needs, with an intuitive UI that provide feedback on feasibility of designs.