Smart Prediction

Run Smart Prediction

To run Smart Prediction on an image or multiple images:

In Data, select one or multiple images.
Click Run Smart Prediction.
Choose a model: Panoptic segmentation (trained on COCO), Instance segmentation (trained on COCO), and Semantic segmentation (trained on Cityscapes).
Click Start prediction.

When Smart Prediction completes successfully, you'll see a green checkmark on the Smart Prediction button that corresponds to the image(s). When you hover over that button, you'll see this message: Smart Prediction completed.

When Smart Segmentation is running on an image, the image is inactive. When you hover over it you'll see this message: The item is currently unavailable.

When you run Smart Prediction on an image, open the image in the editor, select the instance you want to annotate, and assign it a class.

Smart Prediction models

Smart Prediction model	Description
Instance segmentation (trained on COCO)	Predicts your instance’s class label, bounding box, and binary mask. Use this model to detect countable objects, such as boxes.
Panoptic segmentation (trained on COCO)	Assigns all the pixels to classes. Use this model to detect countable objects, such as cars, and uncountable objects, such as the sky.
Semantic segmentation (trained on Cityscapes)	Assigns all the pixels to classes. Use this model to annotate street scenes.