GPT-6 Astra | How to label images automatically and train a specialized vision model

In this video, Lucas Fochesatto, Full Stack Developer at Roboflow, shows how to auto label an object detection dataset using GPT-6 Astra, review the results, and train an specialized vision model that runs at production speed. The example dataset shows a retail shelf of medicine boxes and bottles, and the goal is detecting only the products in the very front row. Each single frame holds around 90 objects, so labeling 60 images by hand means drawing thousands of boxes. Auto Label replaces that with two shorter steps: the frontier model draws the first pass, and you review it! Why do that? Instead of running an extremely large frontier model on every frame in production, we only need GPT-6 Astra to create the labels once. From there, a human corrects them, and those labels train a small specialized model that runs efficiently in your environment. Chapters: 0:00 Introduction: Auto Labeling With Frontier Models 0:21 What are we building? Detecting Items on the Front Row 0:47 Labeling by Hand Does Not Scale 3:05 Tutorial: Setting Up Auto Label in Roboflow 3:41 Writing Class Descriptions That Are Specific 4:43 Testing Gemini vs GPT-6 Astra on the Same Images 6:00 Running the Job and Reviewing the Labels 7:38 Training an RF-DETR Model on the Dataset 9:25 Deploying the Model to an Application = Resources = Auto Label with GPT-6 Astra: https://blog.roboflow.com/auto-label-with-gpt-6-astra/ Auto Label documentation: https://docs.roboflow.com/datasets/annotate/annotate/ai-labeling/auto-label GPT-6 Astra on the Model Playground: https://playground.roboflow.com/models/openai/gpt-6-astra Join a future live session: https://roboflow.com/webinar