Acceleration of ResNet18 Based on Run-time Inference Engine

ResNet18 is a deep convolutional neural network that belongs to the ResNet (Residual Network) family. Due to its excellent performance and relatively shallow network depth, it has a wide and flexible range of applications, particularly excelling in scenarios where computational resources are limited but high precision is required. It is widely used in image classification, object detection, and real-time applications in computer vision. This paper implements the acceleration of the ResNet18 network using an inference engine and compares the results. Experimental result shows that when using the inference engine for acceleration, inference speed is significantly improved in all precision levels, with particularly notable improvements in INT8 precision.