src.tests.spconv3d_decoding_layer_test

Classes

SpConv3DDecodingLayerTest()

class src.tests.spconv3d_decoding_layer_test.SpConv3DDecodingLayerTest

Author:

Alberto M. Esmoris Pena

Test for the SpConv decoder layer. Verifies that the decoder consumes the four-element input list [F, skip_link, S, U] and emits a tensor of the expected shape (1 + R_t, n_{fd}). Also exercises:

• Each residual strategy (None, "sharedmlp", "ssc3d").

• The ground row invariant on the output.

• The uW_regularizer / uW_constraint kwargs are honoured (regularizer instance is attached, not silently dropped).

• get_config serializes the skip-link kernel correctly: round-trip a decoder’s config and ensure all three skip-link keys carry the correct dict types after re-deserialization.

__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_shapes_no_residual()

subtest_shapes_residual_sharedmlp()

subtest_shapes_residual_ssc3d()

subtest_ground_row_invariant()

subtest_uW_regularizer_attached()

uW_regularizer must end up as a real regularizer instance.

subtest_get_config_serializers()

get_config must serialize the skip-link kernel’s initializer / constraint with the correct Keras modules. Verify that the serialized dict keys round-trip through the right deserialize calls. (Not reconstructing the full layer here — just checking that the serializers produce dicts shaped like what initializers / constraints emit.)