The use of "default_class_id" in multiclass classification

[greenhawktobias]

Hi,
I am doing a test with some images of mowed grass and uncut grass on drone images using raster vision Release 0.13.0
I was thinking about the class ids to use.
I have defined the classes in my geojson label file (image)
Background is set to class_id=2, but is it correct to set default_class_id=0 in this case?

Another question: Does chipsize affect the classification results? As in this case with a images resolution of 2 cm
and mowed grass area of 10*8 meters maybe smaller chipsize is beneficial?

CLASS_NAMES = [
    'mowed', 'uncut','background'
]
CLASS_COLORS = [
    '#ffff00', '#0000ff', '#00ffff'

default_class_id=0

rasterizer_config=RasterizerConfig(background_class_id=2)))

[AdeelH]

default_class_id is for GeoJSON features that don't have any class info. If all the features in your GeoJSON have a class ID or name, then default_class_id will have no effect. If you set it to None, features without any class info will be dropped.

See docs. These docs are not for v0.13 but the class inference logic is the same.

default_class_id is mostly useful when you have just one foreground class and all the GeoJSON features belong to that class but don't have class info in their properties. In such a case, default_class_id saves you the trouble of altering your GeoJSON file,