# Train RF-DETR Segmentation (Preview) Models in Roboflow Train

<figure><img src="https://2667452268-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMR3m936tBXGm5QsAcPwe%2Fuploads%2FdsXPH9UVGEiSz7MuT8TW%2FScreenshot%202025-10-02%20at%2022.16.49.png?alt=media&#x26;token=c9e176a6-76ec-4904-a90b-59847191cdd0" alt=""><figcaption></figcaption></figure>

On October 2nd 2025, we launched a preview for [RF-DETR Segmentation](https://blog.roboflow.com/train-rf-detr-segmentation/), a new, state-of-the-art instance segmentation model architecture.

You can now train RF-DETR Segmentation (Preview) models in Roboflow Train. You must have an Instance Segmentation project to use RF-DETR Segmentation training. Semantic segmentation is not supported.

To train an RF-DETR Segmentation model, create a new dataset version in an Instance Segmentation project. Then, follow our guide to train a model with Roboflow Train. Select RF-DETR as the model architecture to use in your training job.

To learn more about RF-DETR Segmentation, [read the model launch announcement](https://blog.roboflow.com/train-rf-detr-segmentation/).