Benchmark neural network model

Overview

The platform must be able to evaluate the performance of the trained models. This can be done by calculating an application score, which can depend on the task solved by the model or by other measures such as the execution time layer by layer. Some of these functions must be executable on the hardware target!

Compute score metrics

For each of the following task, AIDGE give access to the following metrics:

Classification

• Compute precision/recall

• Compute confusion matrix

Segmentation

• Pixel accuracy

• Compute IoU

Detection

• Compute IoU

Measure excution time

AIDGE can provide the execution time operator wise.