Introduction

This section documents the stand-alone ModelPack usage through Docker.  This allows ModelPack to be trained and deployed without eIQ Portal for advanced cloud-based training pipelines.

For the eIQ Portal version of ModelPack please instead refer to ModelPack with eIQ Portal.

Overview

1. ModelPack Licenses

This section introduces the main concepts around ModelPack license system. Also, it provides the contact information for requesting a new license.

• ModelPack Docker License

2. ModelPack Training System

In this section we explain all the concepts needed to train ModelPack on a custom datasets including datasets formats and data exports to different formats. Model training parameters for basic usage and advanced usage are also included. Finally, an addition tool for logging, capable of visualizing the results in real time, allowing the exploration of the performance of the model after each step/epoch.

• Dataset Format - TFRecord format for ModelPack
• Convert SageMaker Input Manifest file to TFRecord
• ModelPack Training Guide | Detection
• ModelPack Logging System
• ModelPack Training Guide | Advanced Model Tuning

3. Model Evaluation

An validation tool has been built in order to compute the performance of the model in different ways. The tool will allow you to validate the Keras model on a local PC as well as a converted and quantized model for use on remote targets.

• DeepView Validation Tool

4. Model Conversion and Quantization

Everything related to model conversion and quantization is covered in this section. Also, we provide some example commands to transform the model into the desired format.

• ModelPack Quantization and Conversion