Roboflow, an end-to-end computer vision developer platform, and Ultralytics, the creators of the popular open source YOLOv5 model, have partnered to launch a streamlined way to prepare, label, and train computer vision models on custom datasets with an emphasis on active learning
YOLOv5 is a computer vision model in the “You Only Look Once” family of models that achieves high accuracy and speed. The open source model is among the most popular AI models available, attracting over 16,000 GitHub stars from developers. Ultralytics, the model’s author, emphasizes developer ease of use.
Roboflow‘s developer tools enable teams of varying machine learning skill to create accurate computer vision models quickly without reinventing the wheel on infrastructure. That includes providing tools that optimize the machine learning operations workflow like dataset management, annotation, model training, deployment, and active learning. To date, over 50,000 developers have built computer vision solutions with Roboflow.
The partnership between Roboflow and Ultraltytics makes training custom YOLOv5 models easier and faster. Users leverage Roboflow to prepare and label custom datasets from images or video, programmatically load them into Ultralytics’s training pipeline, and leverage the Roboflow pip package for automated active learning. This rapidly accelerates the model development lifecycle and enables faster identification of model failure points, which results in better models with even less data.
Products are already being built using Roboflow and Ultralytics. Latest Sightings, a South African company that helps visitors connect with wildlife on safari, leverages Roboflow and YOLOv5 to automatically identify various animal species found in Kruger National Park. Tourists visiting Kruger National Park have a superior experience when they’re able to better understand what animals they saw on their trip and where to go in the National Park to see wildlife.
“At Ultralytics our mission is to make AI easy. We launched our leading Vision AI, YOLOv5, last year, and it has now become one of the world’s most popular AIs due to its simplicity and accuracy in real-world use cases,” explained Glenn Jocher, Ultralytics CEO. “Our new Roboflow integration provides for an even more seamless YOLOv5 training and deployment experience. I’m excited for the possibilities that Ultralytics and Roboflow are opening up through our shared vision of easy AI, and can’t wait to see what our users create using our new integration.”
From the day of it’s June 2020 release, Roboflow has been creating educational content on computer vision models including how to train custom YOLOv5 detectors. “The number of computer vision models that improve performance and advance state of the art is growing. What is especially notable about the Ultralytics team is their emphasis on continuously incorporating improvements into models like YOLOv5,” explained Joseph Nelson, Roboflow CEO. “This shared focus on developer experience is a key component in our mission of democratizing computer vision. Having a tighter integration with leading models like YOLOv5 enables developers to transform every industry, like producing products with fewer errors, interpreting imagery in healthcare, and adding new interactive elements to physical board games. We’re thrilled to see what developers build now that their software has the sense of sight.”