Usage (GLCM)¶

Given that ar is an np.ndarray.

OOP GLCM Transform

>>> from glcm_cupy import GLCM
>>> g = GLCM(...).run(ar)

Functional GLCM Transform

>>> from glcm_cupy import glcm
>>> g = glcm(ar, ...)

I/O¶

For a np.ndarray or cp.ndarray input, the algorithm will output a np.ndarray or cp.ndarray respectively.

Input Array Requirements¶

Shape: \(B\times H\times W\times C\) or \(H\times W\times C\)
Non-negative, integer array
Must be large enough for radius and step_size

\[\begin{split}B: \text{Batches}\\ H: \text{Height}\\ W: \text{Width}\\ C: \text{Channel}\end{split}\]

Output Array¶

Shape: \(B\times H^*\times W^*\times C\times F\) or \(H\times W\times C\times F\)
0 to 1 if normalize_features, else it depends on feature.

\[\begin{split}H^*: H - (\text{Step Size} + \text{Radius}) \times 2\\ W^*: W - (\text{Step Size} + \text{Radius}) \times 2\\ F: \text{Features}\end{split}\]

Arguments¶

Argument	Description	Default
directions	List of `GLCM.Direction` for GLCM to use	`(Direction.EAST, Direction....)`
features	List of `GLCM.Features` for GLCM to use	`(Features.HOMOGENEITY, Features....)`
step_size	Distance between GLCM Windows	`1`
radius	Radius of GLCM Windows	`2`
bin_from	Maximum Value + 1 of the array.	`256`
bin_to	Maximum Value + 1 of the array	`256`
normalized_features	Whether to scale features to `[0, 1]`	`True`
verbose	Whether `tqdm` outputs progress	`True`
max_partition_size1	No. of windows parsed per GLCM Matrix	`10000`
max_threads1	No. of threads per CUDA block	`512`