As GPU is a primary target for the CNN acceleration with Intel® OpenVX* implementation, you should try the resulting app on the GPU.

The easiest way is to play with the system-level device affinities, as explained in the System-Level Device Affinities topic of Intel® Computer Vision SDK Developer Guide.

If you want to schedule the graph programmatically, refer to the Scheduling Individual Nodes to Different Targets topic of the Intel® CV SDK Developer Guide and SDK samples that feature heterogeneous execution.

Nota de otimização: 
Para obter informações mais completas sobre otimizações do compilador, consulte nosso aviso de otimização.