Training a CNN model for real world computer vision tasks on the CPU is not feasible due to the high constraint imposed by the CNN architecture.

A single class (i.e. face) CNN model would take one year to complete training on a standard Intel desktop CPU.

A GPU capable special SOD release which is not available to the public is required in order to train your own CNN model. Visit the download page for additional information or contact support for inquiry about the GPU capable SOD.

For CPU capable training, we recommend that you rely on the RealNets model training which is reasonably fast (16 hours to few days to complete) on a modern desktop CPU.