available_separations

available_separations()

The separation signals you can name under separation, and what each one costs.

instance_segmentation only, and one block per model rather than a list: this is what the network is trained to predict, so it sets the width of the head and which losses are admissible, and several targets do not share a training run.

Returns

Name	Type	Description
	list[dict[str, Any]]	One row per signal: the name, the arguments its block accepts, how wide a head it
	list[dict[str, Any]]	needs at n_class 2, and what it predicts. The widths differ by where the extra
	list[dict[str, Any]]	channels go - rim adds a third class, flows adds two continuous channels in
	list[dict[str, Any]]	front of the classes - which is why nothing states a head width in a file.

[row[“name”] for row in available_separations()] [‘flows’, ‘rim’] next(r for r in available_separations() if r[“name”] == “rim”)[“arguments”] (‘width’,) next(r for r in available_separations() if r[“name”] == “flows”)[“arguments”] (‘mode’, ‘loss’, ‘weight’) [(r[“name”], r[“head_channels”]) for r in available_separations()] [(‘flows’, 4), (‘rim’, 3)]