IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[Common] Unique decomposition: fix the single element case (#13918)

Browse Source 
Signed-off-by: Kazantsev, Roman <roman.kazantsev@intel.com>

Signed-off-by: Kazantsev, Roman <roman.kazantsev@intel.com>
This commit is contained in:
Roman Kazantsev
2022-11-10 01:25:43 +03:00
committed by GitHub
parent c7cd2645f4
commit aaefe0a256
2 changed files with 22 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
src/common/transformations/src/transformations/op_conversions/unique_decomposition.cpp
View File

@@ -69,6 +69,27 @@ ov::pass::UniqueDecomposition::UniqueDecomposition() {
return false;
}
// in case the input with the single input, there is no transformation required
// this separate pass needed since now there is no handler for empty tensors
// that should be converted to empty Constant nodes
// x is already of rank equal to 1 after Reshape
auto x_shape = x->get_input_partial_shape(0);
if (x_shape[0].is_static() && x_shape[0] == 1) {
if (!unique_node->get_output_target_inputs(0).empty()) {
x->set_friendly_name(unique_node->get_friendly_name() + ".0");
unique_node->output(0).replace(x->output(0));
}
if (!unique_node->get_output_target_inputs(2).empty()) {
auto zero_const = rg.make<Constant>(output_indices_type, Shape{1}, 0);
zero_const->set_friendly_name(unique_node->get_friendly_name() + ".2");
unique_node->output(2).replace(zero_const->output(0));
}
copy_runtime_info(unique_node, rg.get());
return true;
}
// denote a number of elements in x as n
auto n = get_elements_number_1d(x, element::i32, rg);

3
tests/layer_tests/tensorflow_tests/test_tf_Unique.py
View File

@@ -30,8 +30,7 @@ class TestUnique(CommonTFLayerTest):
return tf_net, None
test_data_basic = [
pytest.param(dict(x_shape=[1], data_type=tf.float32, out_idx=tf.int32),
marks=pytest.mark.xfail(reason="95824")),
dict(x_shape=[1], data_type=tf.float32, out_idx=tf.int32),
dict(x_shape=[50], data_type=tf.float32, out_idx=tf.int32),
dict(x_shape=[100], data_type=tf.float32, out_idx=tf.int64),
]