Edge Developer Toolbox Developer Guide
Create AI Model (Premium Functionality)
This feature is part of the Premium Functionality available to Intel® Tiber™ Edge Platform customers by invitation only. To request access, contact your Intel account manager.
After you have collected relevant data and prepared it, you can start creating AI models. Model building involves selecting an architecture and training the model to identify patterns and relationships in the prepared data.
AI solutions in the field need adjustments as real-world conditions change. For instance, you may have a defect detection model trained to detect a limited number of basic defects. It might later need to identify torn labels or damaged boxes. To address this, the model requires retraining on a new dataset containing images specific to these new use cases. Once retrained, the model undergoes further optimization cycles on your target hardware to meet performance benchmarks before it can be redeployed at scale.
The Edge Developer Toolbox is designed to give businesses this flexibility in the MLOps life cycle. In subsequent sections, we will explore the AI workflow development journey to build and optimize solutions for the edge.
AI Model Development Workflow
The model development workflow is an iterative, experimentation-based process that can span many days or weeks. The two main tasks are:
Train new machine learning (ML) models
Import a dataset of images and an annotation file. The Common Objects in Context (COCO) dataset format is supported.
Train models using the Intel® Geti™ platform, Intel’s computer-vision AI platform.
Assess and improve AI inference performance.
View the model accuracy for different topologies and datasets.
On subsequent days, upload new datasets.
Compare and contrast the accuracy of the different versions of the model.
Follow these sections to build and optimize AI solutions: