src.tests.downsampling_spconv3d_layer_test

Classes

DownsamplingSpConv3DLayerTest()

class src.tests.downsampling_spconv3d_layer_test.DownsamplingSpConv3DLayerTest

Author:

Alberto M. Esmoris Pena

Test for the downsampling SpConv 3D layer.

The downsampling layer is structurally identical to the submanifold layer (a single tf.gather + tf.einsum) but operates on the downsampling table D, mapping depth \(t\) features to depth \(t + 1\). The test mirrors the submanifold suite: numpy reference parity, ground-row invariant, static-method shape.

__init__()

Basic configuration for any VL3D test.

Parameters:

name (str) – Test name

run()

Run the test.

Returns:

True if test is successfully passed, False otherwise.

Return type:

bool

subtest_layer_matches_numpy(wD=2)

subtest_ground_row_invariant()

subtest_static_method_unpadded()

subtest_active_form_random_eq(wD=2)

Active-form variant must be bit-equivalent to pad-based.

subtest_active_form_edge_all_sentinel()

All-zero D — outputs must be all zero in both paths.