object_training_image builtin

object_training_image(
    grid, objects, *, background=0, column="facies", seed=0
)

Categorical training image drawn from channels and ellipsoids.

The image starts as background. Each object set, in order, adds objects of its shape with its code at random places until the code covers its proportion of the cells. A later set overwrites the cells of the earlier ones, so their final shares can end below their proportions; the last object of a set can overshoot. Objects cut by the edges of the grid are as frequent as whole ones.

Every size and angle of a set is a number, or a (min, max) range drawn uniformly for each object. Lengths are in the units of the cell sizes and angles in degrees, in world coordinates: azimuth clockwise from north, dip positive down.

• {"shape": "channel", ...}: a channel across the whole grid, with a flat top and a lens-shaped cross-section, thickness deep at its centreline. Keys width, thickness (optional on a 2D grid), azimuth (default 0), amplitude (how far the centreline swings to each side, default 0 for straight channels) and wavelength (distance between two bends on the same side, needed with amplitude). On a 2D grid, a sinuous band.
• {"shape": "ellipsoid", ...}: keys radii (major, semi-major, minor; the minor radius is optional on a 2D grid), azimuth, dip and rake of the major axis (default 0). On a 2D grid, an ellipse along azimuth.

Every set also has code, an integer 0 to 254, and proportion, above 0 and below 1.

Parameters:

Name	Type	Description	Default
grid	BlockModel	The grid whose geometry the image takes, 2D (one layer) or 3D, optionally rotated.	required
objects	sequence of dict	The object sets, in the order they are laid down.	required
background	int	Code of the cells no object covers, 0 to 254.	0
column	str	Name of the code column.	"facies"
seed	int	The same seed and objects give the same image on any number of threads.	0

Returns:

Type	Description
BlockModel	A regular model on the geometry of grid with the codes in column.

Raises:

Type	Description
InvalidInput	If a set has an unknown shape or key, a size is not positive, a range is reversed, a proportion is outside (0, 1), or a set has not reached its proportion after 100 000 objects.

Examples:

>>> grid = bt.BlockModel((0, 0, 0), (1, 1, 1), (120, 120, 1))
>>> ti = bt.object_training_image(
...     grid,
...     [
...         {"shape": "channel", "code": 1, "proportion": 0.25, "width": 8,
...          "azimuth": (-20, 20), "amplitude": 8, "wavelength": 60},
...         {"shape": "ellipsoid", "code": 2, "proportion": 0.05, "radii": (10, 6)},
...     ],
...     seed=0,
... )

used in

• several training images
• training images and their consistency
• seismic volumes