Keylabs 2.3-1: Support for RoboFlow and Custom Yolo Models
We are excited to announce new Keylabs 2.3-1 features that revolutionize semi-automated annotation. This version's main advantage is the integration with RoboFlow and support for custom Yolo models for assisted data annotation.
Initial integration with RoboFlow
You can now run a local RoboFlow inference node on the Keylabs platform. This makes it easier to recognize objects using ML assisted annotation. We have implemented a simplified interface on top of the SDK, allowing you to run models directly on this node. This will allow you to use RoboFlow in your workflow without having to switch to external platforms.
Integration of custom Yolo models
Keylabs can now use Yolo models that users have trained and uploaded to RoboFlow. You can apply these models directly in the Keylabs annotation editor. This allows you to automatically predict objects in images and then quickly manually edit, verify, or confirm them.
This speeds up the data annotation process, reduces the workload on the annotator teams, allows them to focus on more complex tasks, and provides better accuracy of the data input to the ML model.
How it works in Keylabs:
- The user trains a model in RoboFlow and deploys it to the inference node.
- Keylabs communicates with this node in order to send images and video frames to the model.
- The model performs the predictions, finds objects, their coordinates, classes, etc.
- Keylabs displays the results in the annotation editor. Users see the predicted objects and can confirm or edit them.
Keylabs 2.3-1 User Benefits
- Works with your models directly on the Keylabs platform.
- The process takes place locally, without delays in transferring data to the cloud.
- Safe because the data does not leave the local network, which reduces the risk of information leak.
Ready to try new features?
Connect your RoboFlow node and test your own Yolo model right now. Keylabs 2.3-1 is not just an annotation tool. It is a platform that understands and adapts to your machine-learning needs.
