Launch: Train Larger Models on Roboflow

Roboflow has announced the expansion of its Roboflow Train service to support larger model sizes—Medium, Large, and Extra Large—within the Roboflow 3.0 architecture, aimed at enhancing the accuracy of computer vision tasks such as object detection, instance segmentation, and keypoint detection. This upgrade is available to customers on the Basic and Growth plans, allowing them to train more accurate models in the cloud, which, despite running slower than the default Fast model architecture, offer improved performance for projects where precision is critical. Users can initiate the training process by selecting a model architecture, choosing a checkpoint—such as the Microsoft COCO checkpoint for initial training—and monitoring the training progress through live graphs. This upgrade is part of an effort to provide robust solutions for complex computer vision applications, with comprehensive documentation available for users seeking to maximize the utility of Roboflow Train.