def forward(self, input, expand_ratio=None, channel=None):
self.cur_config = {'expand_ratio': expand_ratio, 'channel': channel}
### weight: (Cin, Cout)
in_nc = int(input.shape[-1])
assert (
expand_ratio == None or channel == None
), "expand_ratio and channel CANNOT be NOT None at the same time."
if expand_ratio != None:
out_nc = int(expand_ratio * self.base_output_dim)
elif channel != None:
out_nc = int(channel)
else:
out_nc = self.output_dim
weight = self.weight[:in_nc, :out_nc]
if self._bias_attr != False:
bias = self.bias[:out_nc]
use_bias = True
pre_bias = _varbase_creator(dtype=input.dtype)
core.ops.matmul(input, weight, pre_bias, 'transpose_X', False,
'transpose_Y', False, "alpha", 1)
if self._bias_attr != False:
pre_act = dygraph_utils._append_bias_in_dygraph(
pre_bias, bias, axis=len(input.shape) - 1)
else:
pre_act = pre_bias
return dygraph_utils._append_activation_in_dygraph(pre_act, self._act)
def forward(self, input, config):
in_nc = int(input.shape[1])
out_nc = int(config['channel'])
weight = self.weight[:in_nc, :out_nc, :, :]
if in_dygraph_mode():
op = getattr(core.ops, self._op_type)
out = op(input, weight, 'output_size', self._output_size,
'strides', self._stride, 'paddings', self._padding,
'dilations', self._dilation, 'groups', self._groups,
'use_cudnn', self._use_cudnn)
pre_bias = out
if self.bias is not None:
bias = self.bias[:out_nc]
pre_act = dygraph_utils._append_bias_in_dygraph(
pre_bias, bias, 1)
else:
pre_act = pre_bias
return dygraph_utils._append_activation_in_dygraph(pre_act,
act=self._act)
check_variable_and_dtype(input, 'input',
['float16', 'float32', 'float64'],
"SuperConv2DTranspose")
inputs = {'Input': [input], 'Filter': [weight]}
attrs = {
'output_size': self._output_size,
'strides': self._stride,
'paddings': self._padding,
'dilations': self._dilation,
'groups': self._groups,
'use_cudnn': self._use_cudnn
}
pre_bias = self._helper.create_variable_for_type_inference(
dtype=input.dtype)
self._helper.append_op(type=self._op_type,
inputs=inputs,
outputs={'Output': pre_bias},
attrs=attrs)
if self.bias is not None:
pre_act = self._helper.create_variable_for_type_inference(
dtype=self._dtype)
self._helper.append_op(type='elementwise_add',
inputs={
'X': [pre_bias],
'Y': [bias]
},
outputs={'Out': [pre_act]},
attrs={'axis': 1})
else:
pre_act = pre_bias
out = self._helper.append_activation(pre_act, act=self._act)
return out
def forward(self,
input,
kernel_size=None,
expand_ratio=None,
channel=None):
self.cur_config = {
'kernel_size': kernel_size,
'expand_ratio': expand_ratio,
'channel': channel
}
in_nc = int(input.shape[1])
assert (
expand_ratio == None or channel == None
), "expand_ratio and channel CANNOT be NOT None at the same time."
if expand_ratio != None:
out_nc = int(expand_ratio * self.base_channel)
elif channel != None:
out_nc = int(channel)
else:
out_nc = self._num_filters
ks = int(
self._filter_size[0]) if kernel_size == None else int(kernel_size)
groups, weight_in_nc, weight_out_nc = self.get_groups_in_out_nc(
in_nc, out_nc)
weight = self.get_active_filter(weight_in_nc, weight_out_nc, ks)
if kernel_size != None or 'kernel_size' in self.candidate_config.keys(
):
padding = convert_to_list(get_same_padding(ks), 2)
else:
padding = self._padding
if self._l_type == 'conv2d':
attrs = ('strides', self._stride, 'paddings', padding, 'dilations',
self._dilation, 'groups', groups if groups else 1,
'use_cudnn', self._use_cudnn)
out = core.ops.conv2d(input, weight, *attrs)
elif self._l_type == 'depthwise_conv2d':
attrs = ('strides', self._stride, 'paddings', padding, 'dilations',
self._dilation, 'groups',
groups if groups else self._groups, 'use_cudnn',
self._use_cudnn)
out = core.ops.depthwise_conv2d(input, weight, *attrs)
else:
raise ValueError("conv type error")
pre_bias = out
out_nc = int(pre_bias.shape[1])
if self.bias is not None:
bias = self.bias[:out_nc]
pre_act = dygraph_utils._append_bias_in_dygraph(pre_bias, bias, 1)
else:
pre_act = pre_bias
return dygraph_utils._append_activation_in_dygraph(pre_act, self._act)
def forward(self,
input,
kernel_size=None,
expand_ratio=None,
channel=None):
if not in_dygraph_mode():
_logger.error("NOT support static graph")
self.cur_config = {
'kernel_size': kernel_size,
'expand_ratio': expand_ratio,
'channel': channel
}
in_nc = int(input.shape[1])
assert (
expand_ratio == None or channel == None
), "expand_ratio and channel CANNOT be NOT None at the same time."
if expand_ratio != None:
out_nc = int(expand_ratio * self.base_channel)
elif channel != None:
out_nc = int(channel)
else:
out_nc = self._num_filters
ks = int(
self._filter_size[0]) if kernel_size == None else int(kernel_size)
groups, weight_in_nc, weight_out_nc = self.get_groups_in_out_nc(
in_nc, out_nc)
weight = self.get_active_filter(weight_in_nc, weight_out_nc, ks)
if kernel_size != None or 'kernel_size' in self.candidate_config.keys(
):
padding = convert_to_list(get_same_padding(ks), 2)
else:
padding = self._padding
op = getattr(core.ops, self._op_type)
out = op(input, weight, 'output_size', self._output_size, 'strides',
self._stride, 'paddings', padding, 'dilations',
self._dilation, 'groups', groups, 'use_cudnn',
self._use_cudnn)
pre_bias = out
out_nc = int(pre_bias.shape[1])
if self.bias is not None:
bias = self.bias[:out_nc]
pre_act = dygraph_utils._append_bias_in_dygraph(pre_bias, bias, 1)
else:
pre_act = pre_bias
return dygraph_utils._append_activation_in_dygraph(pre_act,
act=self._act)
def forward(self, input):
quant_input = self._fake_quant_input(input)
quant_weight = self._fake_quant_weight(self.weight)
if in_dygraph_mode() and self._l_type == 'conv2d':
attrs = ('strides', self._stride, 'paddings', self._padding,
'dilations', self._dilation, 'groups',
self._groups if self._groups else 1, 'use_cudnn',
self._use_cudnn)
pre_bias = core.ops.conv2d(quant_input, quant_weight, *attrs)
pre_act = dygraph_utils._append_bias_in_dygraph(
pre_bias, self.bias, 1)
return dygraph_utils._append_activation_in_dygraph(
pre_act, self._act)
check_variable_and_dtype(quant_input, 'input',
['float16', 'float32', 'float64'],
'QuantizedConv2D')
attrs = {
'strides': self._stride,
'paddings': self._padding,
'dilations': self._dilation,
'groups': self._groups if self._groups else 1,
'use_cudnn': self._use_cudnn,
'use_mkldnn': False,
}
pre_bias = self._helper.create_variable_for_type_inference(
dtype=self._dtype)
self._helper.append_op(type=self._l_type,
inputs={
'Input': quant_input,
'Filter': quant_weight,
},
outputs={"Output": pre_bias},
attrs=attrs)
if self.bias is not None:
pre_act = self._helper.create_variable_for_type_inference(
dtype=self._dtype)
self._helper.append_op(type='elementwise_add',
inputs={
'X': [pre_bias],
'Y': [self.bias]
},
outputs={'Out': [pre_act]},
attrs={'axis': 1})
else:
pre_act = pre_bias
return self._helper.append_activation(pre_act, act=self._act)
def forward(self, input):
quant_input = self._fake_quant_input(input)
quant_weight = self._fake_quant_weight(self.weight)
if in_dygraph_mode():
pre_bias = _varbase_creator(dtype=input.dtype)
core.ops.matmul(quant_input, quant_weight, pre_bias, 'transpose_X',
False, 'transpose_Y', False, "alpha", 1)
pre_act = dygraph_utils._append_bias_in_dygraph(
pre_bias, self.bias, axis=len(input.shape) - 1)
return dygraph_utils._append_activation_in_dygraph(
pre_act, self._act)
check_variable_and_dtype(input, 'input',
['float16', 'float32', 'float64'],
"QuantizedLinear")
attrs = {
"transpose_X": False,
"transpose_Y": False,
"alpha": 1,
}
inputs = {"X": [quant_input], "Y": [quant_weight]}
mul_out = self._helper.create_variable_for_type_inference(self._dtype)
self._helper.append_op(type="matmul",
inputs=inputs,
outputs={"Out": [mul_out]},
attrs=attrs)
if self.bias is not None:
pre_activation = self._helper.create_variable_for_type_inference(
dtype=self._dtype)
self._helper.append_op(type='elementwise_add',
inputs={
'X': [mul_out],
'Y': [self.bias]
},
outputs={'Out': [pre_activation]},
attrs={'axis': len(input.shape) - 1})
else:
pre_activation = mul_out
return self._helper.append_activation(pre_activation, act=self._act)
def forward(self, input, expand_ratio=None, channel=None):
self.cur_config = {'expand_ratio': expand_ratio, 'channel': channel}
in_nc = int(input.shape[1])
assert (
expand_ratio == None or channel == None
), "expand_ratio and channel CANNOT be NOT None at the same time."
if expand_ratio != None:
out_nc = int(expand_ratio * self.base_output_dim)
elif channel != None:
out_nc = int(channel)
else:
out_nc = self.conv[0]._num_filters
weight = self.conv[0].weight[:in_nc]
### conv1
if self.conv[0]._l_type == 'conv2d':
attrs = ('strides', self.conv[0]._stride, 'paddings',
self.conv[0]._padding, 'dilations', self.conv[0]._dilation,
'groups', in_nc, 'use_cudnn', self.conv[0]._use_cudnn)
out = core.ops.conv2d(input, weight, *attrs)
elif self.conv[0]._l_type == 'depthwise_conv2d':
attrs = ('strides', self.conv[0]._stride, 'paddings',
self.conv[0]._padding, 'dilations', self.conv[0]._dilation,
'groups', in_nc, 'use_cudnn', self.conv[0]._use_cudnn)
out = core.ops.depthwise_conv2d(input, weight, *attrs)
else:
raise ValueError("conv type error")
pre_bias = out
if self.conv[0].bias is not None:
bias = self.conv[0].bias[:in_nc]
pre_act = dygraph_utils._append_bias_in_dygraph(pre_bias, bias, 1)
else:
pre_act = pre_bias
conv0_out = dygraph_utils._append_activation_in_dygraph(
pre_act, self.conv[0]._act)
norm_out = self.conv[1](conv0_out)
weight = self.conv[2].weight[:out_nc, :in_nc, :, :]
if self.conv[2]._l_type == 'conv2d':
attrs = ('strides', self.conv[2]._stride, 'paddings',
self.conv[2]._padding, 'dilations', self.conv[2]._dilation,
'groups', self.conv[2]._groups if self.conv[2]._groups else
1, 'use_cudnn', self.conv[2]._use_cudnn)
out = core.ops.conv2d(norm_out, weight, *attrs)
elif self.conv[2]._l_type == 'depthwise_conv2d':
attrs = ('strides', self.conv[2]._stride, 'paddings',
self.conv[2]._padding, 'dilations', self.conv[2]._dilation,
'groups', self.conv[2]._groups, 'use_cudnn',
self.conv[2]._use_cudnn)
out = core.ops.depthwise_conv2d(norm_out, weight, *attrs)
else:
raise ValueError("conv type error")
pre_bias = out
if self.conv[2].bias is not None:
bias = self.conv[2].bias[:out_nc]
pre_act = dygraph_utils._append_bias_in_dygraph(pre_bias, bias, 1)
else:
pre_act = pre_bias
conv1_out = dygraph_utils._append_activation_in_dygraph(
pre_act, self.conv[2]._act)
return conv1_out
def forward(self, input, config):
in_nc = int(input.shape[1])
out_nc = config['channel']
weight = self.weight[:out_nc, :in_nc, :, :]
#print('super conv shape', weight.shape)
if in_dygraph_mode():
if self._l_type == 'conv2d':
attrs = ('strides', self._stride, 'paddings', self._padding,
'dilations', self._dilation, 'groups',
self._groups if self._groups else 1, 'use_cudnn',
self._use_cudnn)
out = core.ops.conv2d(input, weight, *attrs)
elif self._l_type == 'depthwise_conv2d':
attrs = ('strides', self._stride, 'paddings', self._padding,
'dilations', self._dilation, 'groups', self._groups,
'use_cudnn', self._use_cudnn)
out = core.ops.depthwise_conv2d(input, weight, *attrs)
else:
raise ValueError("conv type error")
pre_bias = out
if self.bias is not None:
bias = self.bias[:out_nc]
pre_act = dygraph_utils._append_bias_in_dygraph(
pre_bias, bias, 1)
else:
pre_act = pre_bias
return dygraph_utils._append_activation_in_dygraph(
pre_act, self._act)
inputs = {'Input': [input], 'Filter': [weight]}
attrs = {
'strides': self._stride,
'paddings': self._padding,
'dilations': self._dilation,
'groups': self._groups if self._groups else 1,
'use_cudnn': self._use_cudnn,
'use_mkldnn': False,
}
check_variable_and_dtype(input, 'input',
['float16', 'float32', 'float64'],
'SuperConv2D')
pre_bias = self._helper.create_variable_for_type_inference(
dtype=self._dtype)
self._helper.append_op(type=self._l_type,
inputs={
'Input': input,
'Filter': weight,
},
outputs={"Output": pre_bias},
attrs=attrs)
if self.bias is not None:
bias = self.bias[:out_nc]
pre_act = self._helper.create_variable_for_type_inference(
dtype=self._dtype)
self._helper.append_op(type='elementwise_add',
inputs={
'X': [pre_bias],
'Y': [bias]
},
outputs={'Out': [pre_act]},
attrs={'axis': 1})
else:
pre_act = pre_bias
# Currently, we don't support inplace in dygraph mode
return self._helper.append_activation(pre_act, act=self._act)
def forward(self, input, config):
in_nc = int(input.shape[1])
out_nc = int(config['channel'])
weight = self.conv[0].weight[:in_nc]
### conv1
if in_dygraph_mode():
if self.conv[0]._l_type == 'conv2d':
attrs = ('strides', self.conv[0]._stride, 'paddings',
self.conv[0]._padding, 'dilations',
self.conv[0]._dilation, 'groups', in_nc, 'use_cudnn',
self.conv[0]._use_cudnn)
out = core.ops.conv2d(input, weight, *attrs)
elif self.conv[0]._l_type == 'depthwise_conv2d':
attrs = ('strides', self.conv[0]._stride, 'paddings',
self.conv[0]._padding, 'dilations',
self.conv[0]._dilation, 'groups', in_nc, 'use_cudnn',
self.conv[0]._use_cudnn)
out = core.ops.depthwise_conv2d(input, weight, *attrs)
else:
raise ValueError("conv type error")
pre_bias = out
if self.conv[0].bias is not None:
bias = self.conv[0].bias[:in_nc]
pre_act = dygraph_utils._append_bias_in_dygraph(
pre_bias, bias, 1)
else:
pre_act = pre_bias
conv0_out = dygraph_utils._append_activation_in_dygraph(
pre_act, self.conv[0]._act)
norm_out = self.conv[1](conv0_out)
weight = self.conv[2].weight[:out_nc, :in_nc, :, :]
if in_dygraph_mode():
if self.conv[2]._l_type == 'conv2d':
attrs = ('strides', self.conv[2]._stride, 'paddings',
self.conv[2]._padding, 'dilations',
self.conv[2]._dilation, 'groups',
self.conv[2]._groups if self.conv[2]._groups else 1,
'use_cudnn', self.conv[2]._use_cudnn)
out = core.ops.conv2d(norm_out, weight, *attrs)
elif self.conv[2]._l_type == 'depthwise_conv2d':
attrs = ('strides', self.conv[2]._stride, 'paddings',
self.conv[2]._padding, 'dilations',
self.conv[2]._dilation, 'groups',
self.conv[2]._groups, 'use_cudnn',
self.conv[2]._use_cudnn)
out = core.ops.depthwise_conv2d(norm_out, weight, *attrs)
else:
raise ValueError("conv type error")
pre_bias = out
if self.conv[2].bias is not None:
bias = self.conv[2].bias[:out_nc]
pre_act = dygraph_utils._append_bias_in_dygraph(
pre_bias, bias, 1)
else:
pre_act = pre_bias
conv1_out = dygraph_utils._append_activation_in_dygraph(
pre_act, self.conv[2]._act)
return conv1_out
