Evgeny Lazarev
98783b88ac
* Removed transformation which removes Const->Result sub-graphs * Removed one more MO transformation which removes Const->Result sub-graph during the front phase
66 lines
3.2 KiB
Python
66 lines
3.2 KiB
Python
# Copyright (C) 2018-2021 Intel Corporation
|
|
# SPDX-License-Identifier: Apache-2.0
|
|
|
|
from extensions.ops.DetectionOutput import DetectionOutput
|
|
from mo.front.common.partial_infer.utils import int64_array
|
|
from mo.front.subgraph_matcher import SubgraphMatch
|
|
from mo.front.tf.replacement import FrontReplacementFromConfigFileSubGraph
|
|
from mo.graph.graph import Graph
|
|
from mo.ops.const import Const
|
|
from mo.ops.op import PermuteAttrs
|
|
from mo.ops.reshape import Reshape
|
|
from mo.ops.result import Result
|
|
|
|
|
|
class SSDToolboxDetectionOutputReplacement(FrontReplacementFromConfigFileSubGraph):
|
|
replacement_id = 'SSDToolboxDetectionOutput'
|
|
|
|
def nodes_to_remove(self, graph: Graph, match: SubgraphMatch):
|
|
return []
|
|
|
|
def generate_sub_graph(self, graph: Graph, match: SubgraphMatch):
|
|
# IE DetectionOutput layer consumes flattened confidences and locations tensors.
|
|
# That is why we add reshapes before them.
|
|
locs_node = match.single_input_node(0)
|
|
conf_node = match.single_input_node(1)
|
|
prior_boxes_node = match.single_input_node(2)
|
|
|
|
locs_out_nodes = locs_node[0].out_nodes()
|
|
assert len(locs_out_nodes) == 1
|
|
locs_out_node = locs_out_nodes[list(locs_out_nodes.keys())[0]]
|
|
assert locs_out_node.op == "Result", locs_out_node.op
|
|
graph.remove_node(locs_out_node.id)
|
|
|
|
conf_out_nodes = conf_node[0].out_nodes()
|
|
assert len(conf_out_nodes) == 1
|
|
conf_out_node = conf_out_nodes[list(conf_out_nodes.keys())[0]]
|
|
assert conf_out_node.op == "Result", conf_out_node.op
|
|
graph.remove_node(conf_out_node.id)
|
|
|
|
# reshape operation to flatten confidence tensor
|
|
const = Const(graph, {'value': int64_array([0, -1])}).create_node()
|
|
reshape_loc_node = Reshape(graph, {}).create_node([locs_node, const], dict(name='DetectionOutput_Reshape_loc_'))
|
|
|
|
# reshape operation to flatten confidence tensor
|
|
reshape_conf_node = Reshape(graph, {}).create_node([conf_node, const], dict(name='DetectionOutput_Reshape_conf_'))
|
|
|
|
# remove the Result node after the priors node
|
|
assert prior_boxes_node[0].out_node().op == "Result"
|
|
graph.remove_node(prior_boxes_node[0].out_node().id)
|
|
|
|
# reshape operation for prior boxes tensor
|
|
const = Const(graph, {'value': int64_array([1, 2, -1])}).create_node()
|
|
reshape_priors_node = Reshape(graph, {}).create_node([prior_boxes_node, const],
|
|
dict(name='DetectionOutput_Reshape_priors_'))
|
|
# create Detection Output node with three inputs: locations, confidences and prior boxes
|
|
detection_output_op = DetectionOutput(graph, match.custom_replacement_desc.custom_attributes)
|
|
detection_output_node = detection_output_op.create_node(
|
|
[reshape_loc_node, reshape_conf_node, reshape_priors_node],
|
|
dict(name=detection_output_op.attrs['type'] + '_'))
|
|
PermuteAttrs.set_permutation(reshape_priors_node, detection_output_node, None)
|
|
|
|
# create Output node to mark DetectionOutput as a graph output operation
|
|
output_op = Result(graph)
|
|
output_op.create_node([detection_output_node], dict(name='sink_'))
|
|
return {}
|