ocnn.dataset

Transform	A boilerplate class which transforms an input data for ocnn.
CollateBatch	Merge a list of octrees and points into a batch.

class Transform(depth: int, full_depth: int, distort: bool, angle: list, interval: list, scale: float, uniform: bool, jitter: float, flip: list, orient_normal: str = '', **kwargs)

A boilerplate class which transforms an input data for ocnn. The input data is first converted to Points, then randomly transformed (if enabled), and converted to an Octree.

Parameters:

• depth (int) – The octree depth.

• full_depth (int) – The octree layers with a depth small than full_depth are forced to be full.

• distort (bool) – If true, performs the data augmentation.

• angle (list) – A list of 3 float values to generate random rotation angles.

• interval (list) – A list of 3 float values to represent the interval of rotation angles.

• scale (float) – The maximum relative scale factor.

• uniform (bool) – If true, performs uniform scaling.

• jitter (float) – The maximum jitter values.

• flip (list) – A list of 3 float values to represent the probability of flipping each axis.

• orient_normal (str) – Orient point normals along the specified axis, which is useful when normals are not oriented.

preprocess(sample: dict, idx: int)

Transforms sample to Points and performs some specific transformations, like normalization.

transform(sample: dict, idx: int)

Applies the general transformations provided by ocnn.

points2octree(points: Points)

Converts the input points to an octree.

rnd_parameters()

Generates random parameters for data augmentation.

class CollateBatch(merge_points: bool = False)

Merge a list of octrees and points into a batch.