Abstract: Various techniques are disclosed to provide for automated verification of the performance of embedded artificial neural networks (ANNs). In one example, a method includes converting a reference ANN to generate an embedded ANN for deployment on an imaging device. The method also includes deploying the reference ANN on a host. The method also includes processing predetermined images by the reference ANN on the host to generate host inference results. The method also includes receiving device inference results from the imaging device at the host in response to processing of the predetermined images by the embedded ANN on the imaging device. The method also includes comparing the device inference results with the host inference results to evaluate performance of the embedded ANN in relation to the reference ANN. Additional methods, devices, and systems are also provided.