Train and use YOLO26 semantic segmentation models

You can now label, train, and deploy YOLO26 semantic segmentation models in Roboflow. Create a semantic segmentation project, choose YOLO26 as the architecture, and train on Roboflow's hosted GPUs. In addition, cityscapes-pretrained checkpoints are available as public pre-trained models for you to use in Workflows.

Semantic segmentation assigns a class to every pixel, producing a dense map of the whole scene rather than boxes or per-instance masks, which suits scene-level tasks like autonomous driving, land-cover mapping, and medical imaging. Once trained, you can test the model in the web interface, deploy it through the cloud API or on your own hardware with Inference, and chain it into Workflows.

Read the announcement post