openvino/model-optimizer/extensions/ops/correlation.py

67 lines
1.9 KiB
Python

# Copyright (C) 2018-2021 Intel Corporation
# SPDX-License-Identifier: Apache-2.0
from math import ceil
# Concat infer : N - number of inputs to concat
# axis - dimension number for tensors concatenation
import numpy as np
from mo.graph.graph import Node, Graph
from mo.ops.op import Op
class CorrelationOp(Op):
op = 'Correlation'
def __init__(self, graph: Graph, attrs: dict):
mandatory_props = {
'type': __class__.op,
'op': __class__.op,
'version': 'extension',
'in_ports_count': 1,
'out_ports_count': 1,
'infer': CorrelationOp.corr_infer
}
super().__init__(graph, mandatory_props, attrs)
def supported_attrs(self):
return [
'pad',
'kernel_size',
'max_displacement',
'stride_1',
'stride_2',
'single_direction',
'do_abs',
'correlation_type'
]
@staticmethod
def corr_infer(node: Node):
outn = node.out_node(0)
inn = node.in_node(0)
outn.shape = np.zeros(4, dtype=int)
outn.shape[0] = inn.shape[0]
bottomchannels = inn.shape[1]
paddedbottomheight = inn.shape[2]
paddedbottomwidth = inn.shape[3] + 2 * node.pad
kernel_radius_ = (node.kernel_size - 1) / 2;
border_size_ = node.max_displacement + kernel_radius_
outn.shape[3] = ceil((float)(paddedbottomwidth - border_size_ * 2) / node.stride_1)
outn.shape[2] = ceil((float)(paddedbottomheight - kernel_radius_ * 2) / node.stride_1)
neighborhood_grid_radius_ = node.max_displacement / node.stride_2
if node.single_direction != 0:
neighborhood_grid_width_ = neighborhood_grid_radius_ + 1
else:
neighborhood_grid_width_ = neighborhood_grid_radius_ * 2 + 1
outn.shape[1] = neighborhood_grid_width_ * neighborhood_grid_width_