Analysis of the Acceleration of Neural Networks Inference on Intel Processors Based on OpenVINO Toolkit

The article studies the performance of a trained neural network SSD_MobileNet_V2_COCO. It is proposed to use the OpenVINO Toolkit to increase network performance. Performance evaluation is calculated by the reciprocal of the frame processing time, which characterizes the number of frames processed per second. Dataset COCO (by Microsoft) was used as the source dataset. In this case, 200 images were selected from this dataset, and during processing all images were reduced to the same sizes 300x300. Studies shown that the use of OpenVINO has increased the performance of the neural network SSD_MobileNet_V2_COCO by 130 times on average. At the same time, in contrast to starting a network using TensorFlow only, the variance of network performance using OpenVINO is significantly increased. However, the use of such an accelerator remains appropriate on Intel processors.