predicting a compensated model based on a three-dimensional (3D) object model having global values and local values for each of a plurality of edges, wherein the compensated model comprises of a training object model point cloud representing the 3D object model with modifications that compensate for a predicted manufacturing deformation, wherein the global values provide global information for simulating a thermal mass effect feature of the plurality of edges, and wherein the local values provide local neighborhood information for simulating a thermal diffusion effect feature of the plurality of edges;

predicting a deformed model based on convolution of features of the plurality of edges for the compensated model predicted;

comparing the deformed model to the 3D object model; and

based on comparing the deformed model to the 3D object model, adjusting one or more of the 3D object model, the compensation machine learning model, the compensated model, or a printing variable to reduce geometric inaccuracy in a 3D object to be printed.