Uploaded March 2026 | Updated November 2026, 16 hours ago
In this video, Carmen Lee, Product Designer at Roboflow, walks us through how to improve vision model performance with continuous improvement loops in Roboflow Rapid.
Roboflow Rapid is designed to get your custom computer vision model built and ready to deploy in minutes using text prompts and foundation models. From there, Rapid can distill those automated labels by fine-tuning a much lighter, custom vision model that is built for speed and efficiency.
But the process doesn't stop at deployment. Carmen demonstrates how to combine Rapid with Roboflow Workflows to bring new, real-world data back into your training set, creating an automated active learning loop.
Using an example visual inspection system designed to detect and count plastic forks in a waste management process, Carmen shows how to funnel production video data directly back into Rapid. From there, you can easily review edge cases, add new classes, and click a single button to retrain and continuously improve your model's accuracy over time!
= Additional Resources =
Roboflow Rapid: rapid.roboflow.com
Roboflow Workflows: roboflow.com/workflows/build
Getting Started Documentation: docs.roboflow.com
= Chapters =
00:00 Introduction to Continuous Improvement
00:41 Use Case: Counting Plastic Forks in Garbage
01:27 Building the Initial Model with Text Prompts
02:37 Deploying the Model Using Roboflow Workflows
03:50 Collecting Real Production Data for Retraining
04:44 Reviewing and Labeling New Frames (Active Learning)
05:31 Comparing Metrics: Analyzing the Improved Model
06:14 The Continuous Improvement Loop
In this video, Carmen Lee, Product Designer at Roboflow, walks us through how to improve vision model performance with continuous improvement loops in Roboflow Rapid.
Roboflow Rapid is designed to get your custom computer vision model built and ready to deploy in minutes using text prompts and foundation models. From there, Rapid can distill those automated labels by fine-tuning a much lighter, custom vision model that is built for speed and efficiency.
But the process doesn't stop at deployment. Carmen demonstrates how to combine Rapid with Roboflow Workflows to bring new, real-world data back into your training set, creating an automated active learning loop.
Using an example visual inspection system designed to detect and count plastic forks in a waste management process, Carmen shows how to funnel production video data directly back into Rapid. From there, you can easily review edge cases, add new classes, and click a single button to retrain and continuously improve your model's accuracy over time!
= Additional Resources =
Roboflow Rapid: rapid.roboflow.com
Roboflow Workflows: roboflow.com/workflows/build
Getting Started Documentation: docs.roboflow.com
= Chapters =
00:00 Introduction to Continuous Improvement
00:41 Use Case: Counting Plastic Forks in Garbage
01:27 Building the Initial Model with Text Prompts
02:37 Deploying the Model Using Roboflow Workflows
03:50 Collecting Real Production Data for Retraining
04:44 Reviewing and Labeling New Frames (Active Learning)
05:31 Comparing Metrics: Analyzing the Improved Model
06:14 The Continuous Improvement Loop










