Class attributes not appearing in instance segmentation prediction results

[kedar]

I’m trying to run predictions on a Roboflow 3.0 Instance Segmentation (Fast) trained model with 5 classifications, including multiple attributes for each class.
e.g. Class 1: Shape, Attributes: Full, Half, etc
Class 2: Defects, Attributes: None, Fungus, Dehydration, etc
Class 3, etc

Each image was labelled with smart polygons to annotate each walnut with multiple classifications and corresponding attributes.

The prediction json for such a model does not seem to return the attributes for each classification or am i missing something in the json?

I was expecting the attributes predicted along with the classes with this instance segmentation approach.

Is this the wrong approach for this problem?

[leo]

Hi @kedar

Annotation attributes are additional pieces of information that are included with the annotation files. They’re designed to be useful for more nuanced data organization, but they are not used during training for our notebooks or Roboflow Train and therefore will not show up in your predictions.