Benchmarking

Purpose

Within the broader context of the project, we are searching for the optimal mappings to distribute the inference of multiple ML models across a heterogeneous device. This work has within this scope, two main objectives that follow the line of benchmarking Google's Coral Edge:

• To split a input tflite model into single operation models, compile and deploy them onto Google's Coral Edge TPU and the CPU, obtaining benchmark times for these inference operations.
• To obtain more precise results regarding a model's inference by analyzing the USB traffic that occurs during its deployment onto Google's Coral Edge TPU.