def common_backpropdata_extender(op: Node):
for attr in [
'strides', 'output_padding', 'pads_begin', 'pads_end', 'dilations'
]:
Extender.attr_to_list(op, attr)
if op.has_valid('output_padding'):
op.output_padding = int64_array([0, 0] + op.output_padding)
dim = len(op.strides)
if op.has_valid('pads_begin') and op.has_valid('pads_end'):
pad = [[0, 0], [0, 0]]
pad.extend([[op.pads_begin[i], op.pads_end[i]] for i in range(dim)])
op['pad'] = int64_array(pad)
op['spatial_dims'] = [i + 2 for i in range(dim)]
if not op.has_valid('dilations'):
op['dilations'] = [1 for _ in range(dim)]
if not op.has_valid('strides'):
op['strides'] = [1 for _ in range(dim)]
op['dilation'] = int64_array([1, 1] + op.dilations)
op['stride'] = int64_array([1, 1] + op.strides)
op['infer'] = backpropdata_infer
def extend(op: Node):
if op.has_valid('classes_index_type'):
op['classes_index_type'] = destination_type_to_np_data_type(
op.classes_index_type)
if op.has_valid('sequence_length_type'):
op['sequence_length_type'] = destination_type_to_np_data_type(
op.sequence_length_type)
def extend(op: Node):
if not op.has_valid('activations'):
op['activations'] = None
mark_input_bins(op, start_port=2)
op['need_copy_input_blobs'] = True
def extend(op: Node):
for attr in [
'strides', 'dilations', 'pads_begin', 'pads_end',
'output_padding'
]:
Extender.attr_to_list(op, attr)
op['stride'] = int64_array([1, 1] + op.strides)
op['dilation'] = int64_array([1, 1] + op.dilations)
op['batch_dims'] = int64_array([0])
op['channel_dims'] = int64_array([1])
if op.has_valid('output_padding'):
op.output_padding = int64_array([0, 0] + op.output_padding)
# Be VERY careful with these attributes!
op['input_feature_channel'] = 1
op['output_feature_channel'] = 0
dim = len(op.pads_begin)
assert dim in (1, 2, 3), '{}D Convolution not supported!'.format(dim)
pad = [[0, 0], [0, 0]]
pad.extend([[op.pads_begin[i], op.pads_end[i]] for i in range(dim)])
op['pad'] = int64_array(pad)
op['spatial_dims'] = [i + 2 for i in range(dim)]
def deconvolution_infer(node: Node):
dims = int64_array(node.in_node(0).shape)
dilations = int64_array(node.dilations)
strides = int64_array(node.strides)
input_n = dims[0]
kernel_shape = int64_array(node.kernel)
kdims = np.where(dilations != 0, (kernel_shape - 1) * dilations + 1,
kernel_shape)
oc = node.output
if node.has_valid('auto_pad') and node.auto_pad in [
'valid', 'same_upper', 'same_lower'
]:
auto_pad = node.auto_pad
if auto_pad == 'valid':
od_temp = (dims[2::] - 1) * strides + kdims
else:
od_temp = dims[2::] * strides
else:
od_temp = strides * (dims[2::] -
1) + kdims - node.pads_begin - node.pads_end
out_shape = [input_n, oc]
for d in od_temp:
out_shape.append(np.int64(d))
node['output_shape'] = int64_array(out_shape)
for n in node.out_nodes():
node.out_node(n).shape = node['output_shape']
def attr_restore(node: Node, attribute: str, value=None):
# Function to restore some specific attr for PriorBox & PriorBoxClustered layers
if not node.has_valid(attribute):
node[attribute] = [] if value is None else [value]
if isinstance(node[attribute], str):
node[attribute] = []
else:
Extender.attr_to_list(node, attribute)
def extend(op: Node):
assert op.has_valid(
'element_type'
), 'Parameter node {} has missed element_type attr!'.format(op.name)
op['data_type'] = destination_type_to_np_data_type(op.element_type)
if op.shape == '':
op.shape = int64_array([])
else:
Extender.attr_to_list(op, 'shape')
def extend(op: Node):
if op.has_valid('output_padding'):
op.output_padding = int64_array([0, 0] + op.output_padding)
dim = len(op.strides)
# assert dim in (2, 3), '{}D Deconvolution not supported!'.format(dim)
if op.has_valid('pads_begin') and op.has_valid('pads_end'):
pad = [[0, 0], [0, 0]]
pad.extend([[op.pads_begin[i], op.pads_end[i]]
for i in range(dim)])
op['pad'] = int64_array(pad)
op['spatial_dims'] = [i + 2 for i in range(dim)]
if not op.has_valid('dilations'):
op['dilations'] = [1 for _ in range(dim)]
if not op.has_valid('strides'):
op['strides'] = [1 for _ in range(dim)]
op['dilation'] = int64_array([1, 1] + op.dilations)
op['stride'] = int64_array([1, 1] + op.strides)
if not op.has_valid('old_type'):
# op['batch_dims'] = int64_array([0]) # ?
op['channel_dims'] = int64_array([1])
op['input_feature_channel'] = 0
op['output_feature_channel'] = 1
op['kernel_spatial'] = op.kernel
# if op.has_valid('auto_pad'):
# op['auto_pad'] = None
op['infer'] = deconvolution_infer # TODO Remove after supporting
else:
op['infer'] = backpropdata_infer
def extend(op: Node):
if op.has_valid('output_type'):
op['output_type'] = destination_type_to_np_data_type(
op.output_type)
def extend(op: Node):
if op.out_port(0).disconnected():
op['remove_values_output'] = True
if op.has_valid('index_element_type'):
op['index_element_type'] = destination_type_to_np_data_type(
op.index_element_type)
def extend(op: Node):
if op.graph.graph['cmd_params'].framework in ('tf', 'caffe'):
op['remove_values_output'] = True
if op.has_valid('index_element_type'):
op['index_element_type'] = destination_type_to_np_data_type(
op.index_element_type)
def extend(op: Node):
if not op.has_valid('activations'):
op['activations'] = None
op['infer'] = Extender.const_shape_infer
def attr_to_list(node: Node, attribute: str):
if not node.has_valid(attribute):
log.warning('Attribute {} missed in node {} with type {}!'.format(
attribute, node.soft_get('name'), node.soft_get('type')))
elif not isinstance(node[attribute], list):
node[attribute] = [node[attribute]]
def extend(op: Node):
if not op.has_valid('activations'):
op['activations'] = None
op['infer'] = Extender.use_shapes_from_ir
def extend(op: Node):
if not op.has_valid('activations'):
op['activations'] = None