`mlvbench.stimuli.shapes`¶

Shape generation utilities.

`random_idsprite(image_size, rng)` ¶

Generate a segmentation map for a random idsprite.

The shape generation is adapted from the idsprites library (Dziadzio et al., 2023).

Parameters:

Name	Type	Description	Default
`image_size`	`int`	Side length of the square image in pixels.	required
`rng`	`Generator`	NumPy random number generator for reproducibility.	required

Returns:

Type	Description
`ndarray`	Region map as uint8 NumPy array of shape `(image_size, image_size, 1)` with values 0 (background) and 1 (foreground shape).

Split the foreground region of a binary segmentation into two sub-regions.

A random line divides the foreground pixels (label 1 in the input) into two sub-regions. Background pixels (label 0) are left unchanged.

Parameters:

Name	Type	Description	Default
`segmentation`	`ndarray`	Binary segmentation map of shape `(H, W, 1)` and dtype `uint8` with values 0 (background) and 1 (foreground). The foreground region will be split into two sub-regions using a random line.	required
`rng`	`Generator`	NumPy random number generator used for reproducibility.	required
`min_region_size`	`float`	Minimum size of each foreground sub-region as a fraction of the total foreground area. Must be in [0.0, 0.5]. When 0.0 (the default) any non-empty split is accepted.	`0.0`

Returns:

Type	Description
`ndarray`	Region map as uint8 numpy array of shape `(H, W, 1)` with values 0 (background), 1 (foreground sub-region A), and 2 (foreground sub-region B).

Raises:

Type	Description
`ValueError`	If `min_region_size` is outside [0.0, 0.5].