Comparing implementations of a Small CNN on Commodity Hardware

Presentation detailing our benchmarking of several low cost boards for AI inference on the Edge.
The NN is a small NN that performs 1 channel images classification over 58 categories targeting an industrial application.
The NN is trained usign TensorflowLite, and quantized to 8-bit.
We perform inference and measure the latency, throughput and power for either CPU, CP+GPU, CPU+TPU mappings.