An Einstein Vision object-detection model must be developed from a training dataset containing representative images of the objects that the model is expected to recognize. Model creation is therefore an implementation activity rather than something that occurs automatically from a single image or during ordinary field execution.
Salesforce ' s Einstein Vision development process begins by collecting representative training images. For object detection, each training example must also identify the relevant objects using bounding-box coordinates. The resulting examples are assembled into a dataset and submitted to the Einstein Vision training service. Training that dataset produces a model that can subsequently identify, classify, count, and locate corresponding objects in new images.
The quality and diversity of the base image set directly influence model performance. A sufficiently representative dataset must account for differences such as product orientation, lighting, shelf arrangement, packaging appearance, and surrounding objects. Salesforce also allows multiple models to be trained from evolving datasets so their performance can be evaluated and refined.
Consequently, the model is deliberately built during implementation from a prepared base dataset before it is incorporated into the field-execution process.
Study Guide Topics:Platform - > Einstein Vision - > Object Detection - > Training Dataset - > Model Creation
===============
