openvino/model-optimizer/extensions/back/TileNormalizer.py

245 lines
8.9 KiB
Python

"""
Copyright (C) 2018-2020 Intel Corporation
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
"""
import numpy as np
from extensions.back.ReshapeMutation import ReshapeMutation
from mo.back.replacement import BackReplacementPattern
from mo.front.common.partial_infer.utils import int64_array
from mo.front.tf.graph_utils import create_op_node_with_second_input
from mo.graph.graph import Graph
from mo.ops.concat import Concat
from mo.ops.const import Const
from mo.ops.squeeze import Squeeze
from mo.ops.tile import Tile, AttributedTile
from mo.ops.unsqueeze import Unsqueeze
from mo.utils.shape import new_shape_node_from_shape_nodes
class TileInputAlignment(BackReplacementPattern):
"""
Aligns rank of data input and length of repeats input of Tile operation
"""
enabled = True
force_clean_up = True
force_shape_inference = True
graph_condition = [lambda graph: not graph.graph['cmd_params'].generate_experimental_IR_V10]
def pattern(self):
return dict(
nodes=[
('tile', dict(kind='op', op='Tile'))
],
edges=[]
)
@staticmethod
def replace_pattern(graph: Graph, match: dict):
node = match['tile']
name = node.soft_get('name', node.id)
input_shape = node.in_port(0).data.get_shape()
assert input_shape is not None
tiles = node.in_port(1).data.get_value()
assert tiles is not None, "Undefined `repeats` (1st port input value) of Tile node '{}'".format(name)
if input_shape.size == tiles.size:
return
if input_shape.size < tiles.size:
unsqueeze = create_op_node_with_second_input(graph, Unsqueeze,
int64_array(list(range(tiles.size - input_shape.size))),
{'name': name + '/input_alignment',
'override_output_shape': True})
node.in_port(0).get_source().connect(unsqueeze.in_port(0))
node.in_port(0).get_connection().set_source(unsqueeze.out_port(0))
else:
const = Const(graph, {'name': name + '/tile_alignment_const',
'value': np.ones([input_shape.size - tiles.size], dtype=np.int64)}).create_node()
concat = Concat(graph, {'axis': 0, 'override_output_shape': True}).create_node()
concat.add_input_port(0)
concat.add_input_port(1)
node.in_port(1).get_source().connect(concat.in_port(1))
node.in_port(1).disconnect()
concat.in_port(0).connect(const.out_port(0))
node.in_port(1).connect(concat.out_port(0))
class Tile3DReshaper(BackReplacementPattern):
enabled = True
force_clean_up = True
def run_before(self):
return [ReshapeMutation, TileVersionDowngrader]
def run_after(self):
return [TileInputAlignment]
@staticmethod
def pattern():
return dict(
nodes=[
('tile', dict(type='Tile'))
],
edges=[]
)
@staticmethod
def replace_pattern(graph: Graph, match: dict):
"""
Workarounds not supported type of Tile in Inference Engine (Tiles are supported for 2-D or 4-D tensors):
Searches for Tiles with 3D shapes and covers it with Reshapes.
Example: Tile (axis=1, tiles=16):
in_shape: [1,1,101]
out_shape: [1,16,101]
Old behaviour:
Tile -> [1,16,101]
New behaviour:
Reshape [1,1,101,1] -> Tile -> [1,16,101,1] -> Reshape [1,16,101]
"""
node = match['tile']
name = node.soft_get('name', node.id)
out_shape = node.out_port(0).data.get_shape()
assert out_shape is not None, 'Output shape is undefined for {} in back phase'.format(name)
if out_shape.size != 3:
return
inp_shape = node.in_port(0).data.get_shape()
assert inp_shape is not None, 'Input shape is undefined for {} in back phase'.format(name)
unsqueeze_dim = Const(graph, {'name': name + '/3D_Tile_Unsqueeze_dim', 'value': int64_array([3])}).create_node()
unsqueeze = Unsqueeze(graph, {'name': name + '/3D_Tile_Unsqueeze', 'override_output_shape': True}).create_node()
unsqueeze_dim.out_port(0).connect(unsqueeze.in_port(1))
const = Const(graph, {'name': name + '/additional_axis', 'value': int64_array([1])}).create_node()
new_tiles = new_shape_node_from_shape_nodes([node.in_port(1).get_source().node, const])
node.in_port(1).get_connection().set_source(new_tiles.out_port(0))
squeeze_dim = Const(graph, {'name': name + '/3D_Tile_Squeeze_dim', 'value': int64_array([3])}).create_node()
squeeze = Squeeze(graph, {'name': name + '/3D_Tile_Squeeze', 'override_output_shape': True}).create_node()
squeeze_dim.out_port(0).connect(squeeze.in_port(1))
source = node.in_port(0).get_source()
node.in_port(0).get_connection().set_source(unsqueeze.out_port(0))
unsqueeze.in_port(0).connect(source)
node.out_port(0).get_connection().set_source(squeeze.out_port(0))
node.out_port(0).connect(squeeze.in_port(0))
node['override_output_shape'] = True
new_tiles['override_output_shape'] = True
node['need_shape_inference'] = True
class TileMultipleAxisReplacer(BackReplacementPattern):
"""
This class convert Tile operation with miltiple != 1 tile dimensions by sequence of Tiles.
"""
enabled = True
force_clean_up = True
graph_condition = [lambda graph: not graph.graph['cmd_params'].generate_experimental_IR_V10]
def run_after(self):
return [TileInputAlignment]
def pattern(self):
return dict(
nodes=[
('tile', dict(kind='op', op='Tile'))
],
edges=[]
)
@staticmethod
def replace_pattern(graph: Graph, match: dict):
tile = match['tile']
name = tile.soft_get('name', tile.id)
tile_array = tile.in_port(1).data.get_value()
assert tile_array is not None, "Undefined `repeats` (1st port input value) of Tile node '{}'".format(name)
assert len(tile_array) == len(tile.in_port(0).data.get_shape())
non_one_tile = np.argwhere(tile_array != 1).flatten()
if len(non_one_tile) == 1:
# We need to add new tiles only in case when we tile more than one dimension
return
if len(non_one_tile) == 0:
# Deleting such Tile that does nothing
tile.out_port(0).get_connection().set_source(tile.in_port(0).get_connection().get_source())
return
last_tile = None
for i in non_one_tile:
tiles = int64_array(np.ones(tile_array.size))
tiles[i] = tile_array[i]
new_tile = create_op_node_with_second_input(graph, Tile, tiles, {'name': name + '/Tile_{}/'.format(i)})
if not last_tile:
last_tile = new_tile
tile.in_port(0).get_connection().set_destination(new_tile.in_port(0))
else:
last_tile.out_port(0).connect(new_tile.in_port(0))
last_tile = new_tile
# Reconnect output to new tile node and delete old tile
tile.out_port(0).get_connection().set_source(last_tile.out_port(0))
class TileVersionDowngrader(BackReplacementPattern):
enabled = True
force_clean_up = True
graph_condition = [lambda graph: not graph.graph['cmd_params'].generate_experimental_IR_V10]
def run_after(self):
return [TileMultipleAxisReplacer]
def pattern(self):
return dict(
nodes=[
('tile', dict(kind='op', op='Tile'))
],
edges=[]
)
@staticmethod
def replace_pattern(graph: Graph, match: dict):
tile = match['tile']
name = tile.soft_get('name', tile.id)
tile_array = tile.in_port(1).data.get_value()
assert tile_array is not None, "Undefined `repeats` (1st port input value) of Tile node '{}'".format(name)
non_one_tile = np.argwhere(tile_array != 1).flatten()
assert len(non_one_tile) == 1
axis = non_one_tile[0]
tiles = tile_array[axis]
new_tile = AttributedTile(graph, {'name': name, 'axis': axis, 'tiles': tiles}).create_node()
tile.out_port(0).get_connection().set_source(new_tile.out_port(0))
tile.in_port(0).get_connection().get_source().connect(new_tile.in_port(0))