def _interpolate(
raw,
input,
size=None,
scale_factor=None,
mode='nearest',
align_corners=None,
):
# 定义的参数包括 scale,即输出与输入的尺寸比例,如 2;scale_h、scale_w,
# 同 scale,分别为 h、w 方向上的尺寸比例;pad_out_h、pad_out_w,仅在 scale 为 2 时
# 有用,对输出进行额外 padding 在 h、w 方向上的数值;upsample_h、upsample_w,输
# 出图像尺寸的数值。在 Upsample 的相关代码中,推荐仅仅使用 upsample_h、
# upsample_w 准确定义 Upsample 层的输出尺寸,其他所有的参数都不推荐继续使用。
# for nearest _interpolate
if mode != "nearest" or align_corners is not None:
raise NotImplementedError("not implement F.interpolate totoaly")
x = raw(input, size, scale_factor, mode)
layer_name = log.add_layer(name='upsample')
top_blobs = log.add_blobs([x], name='upsample_blob'.format(type))
layer = Layer_param(
name=layer_name,
type='Upsample',
bottom=[log.blobs(input)],
top=top_blobs,
)
layer.upsample_param(
size=(input.size(2), input.size(3)),
scale_factor=scale_factor,
)
log.cnet.add_layer(layer)
return x
def _threshold(raw, input, threshold, value, inplace=False):
# for threshold or relu
if value != 0:
raise NotImplementedError("value != 0 not implemented in caffe")
elif value == 0 and threshold == 0:
x = raw(input, threshold, value, inplace)
bottom_blobs = [log.blobs(input)]
layer_name = log.add_layer(name='relu')
log.add_blobs([x], name='relu_blob')
layer = Layer_param(
name=layer_name,
type='ReLU',
bottom=bottom_blobs,
top=[log.blobs(x)],
)
log.cnet.add_layer(layer)
return x
else:
x = raw(input, input, threshold, value, inplace)
bottom_blobs = [log.blobs(input)]
layer_name = log.add_layer(name='threshold')
top_blobs = log.add_blobs([x], name='threshold_blob')
layer = Layer_param(
name=layer_name,
type='Threshold',
bottom=bottom_blobs,
top=top_blobs,
)
layer.param.threshold_param.threshold = threshold
log.cnet.add_layer(layer)
return x
def _conv2d(raw,
input,
weight,
bias=None,
stride=1,
padding=0,
dilation=1,
groups=1):
x = raw(input, weight, bias, stride, padding, dilation, groups)
layer_name = log.add_layer(name='conv')
log.add_blobs([x], name='conv_blob')
layer = Layer_param(
name=layer_name,
type='Convolution',
bottom=[log.blobs(input)],
top=[log.blobs(x)],
)
layer.conv_param(
x.size()[1],
weight.size()[2:],
stride=_pair(stride),
pad=_pair(padding),
dilation=_pair(dilation),
bias_term=bias is not None,
groups=groups,
)
if bias is not None:
layer.add_data(weight.cpu().data.numpy(), bias.cpu().data.numpy())
else:
layer.param.convolution_param.bias_term = False
layer.add_data(weight.cpu().data.numpy())
log.cnet.add_layer(layer)
return x
def _permute(input, *args):
x = raw_permute(input, *args)
layer_name = log.add_layer(name='permute')
log.add_blobs([x], name='permute_blob')
layer = Layer_param(
name=layer_name,
type='Permute',
bottom=[log.blobs(input)],
top=[log.blobs(x)],
)
order1 = args[0]
order2 = args[1]
order3 = args[2]
order4 = args[3]
layer.permute_param(order1, order2, order3, order4)
log.cnet.add_layer(layer)
return x
def _contiguous(input, *args):
x = raw_contiguous(input, *args)
layer_name = log.add_layer(name='contiguous')
log.add_blobs([x], name='contiguous_blob')
layer = Layer_param(
name=layer_name,
type='NeedRemove',
bottom=[log.blobs(input)],
top=[log.blobs(x)],
)
log.cnet.add_layer(layer)
return x
def _sigmoid(raw, input):
# Applies the element-wise function:
# Sigmoid(x)= 1/(1+exp(−x)）
x = raw(input)
layer_name = log.add_layer(name='sigmoid')
log.add_blobs([x], name='sigmoid_blob')
layer = Layer_param(
name=layer_name,
type='Sigmoid',
bottom=[log.blobs(input)],
top=[log.blobs(x)],
)
log.cnet.add_layer(layer)
def _leaky_relu(raw, input, negative_slope=0.01, inplace=False):
x = raw(input, negative_slope)
layer_name = log.add_layer(name='leaky_relu')
log.add_blobs([x], name='leaky_relu_blob')
layer = Layer_param(
name=layer_name,
type='ReLU',
bottom=[log.blobs(input)],
top=[log.blobs(x)],
)
layer.param.relu_param.negative_slope = negative_slope
log.cnet.add_layer(layer)
return x
def _tanh(raw, input):
# for tanh activation
x = raw(input)
layer_name = log.add_layer(name='tanh')
log.add_blobs([x], name='tanh_blob')
layer = Layer_param(
name=layer_name,
type='TanH',
bottom=[log.blobs(input)],
top=[log.blobs(x)],
)
log.cnet.add_layer(layer)
return x
def _relu(raw, input, inplace=False):
# for threshold or prelu
x = raw(input, False)
layer_name = log.add_layer(name='relu')
log.add_blobs([x], name='relu_blob')
layer = Layer_param(
name=layer_name,
type='ReLU',
bottom=[log.blobs(input)],
top=[log.blobs(x)],
)
log.cnet.add_layer(layer)
return x
def _pool(type, raw, input, x, kernel_size, stride, padding, ceil_mode):
# TODO dilation, ceil_mode, return indices
layer_name = log.add_layer(name='{}_pool'.format(type))
top_blobs = log.add_blobs([x], name='{}_pool_blob'.format(type))
layer = Layer_param(
name=layer_name,
type='Pooling',
bottom=[log.blobs(input)],
top=top_blobs,
)
# TODO w,h different kernel, stride and padding
# processing ceil mode
layer.pool_param(
kernel_size=kernel_size,
stride=kernel_size if stride is None else stride,
pad=padding,
type=type.upper(),
ceil_mode=ceil_mode,
)
log.cnet.add_layer(layer)
if ceil_mode is False and stride is not None:
oheight = (input.size()[2] - _pair(kernel_size)[0] +
2 * _pair(padding)[0]) % (_pair(stride)[0])
owidth = (input.size()[3] - _pair(kernel_size)[1] +
2 * _pair(padding)[1]) % (_pair(stride)[1])
if oheight != 0 or owidth != 0:
caffe_out = raw(input,
kernel_size,
stride,
padding,
ceil_mode=True)
print("WARNING: the output shape miss match at {}: "
"input {} output---Pytorch:{}---Caffe:{}\n"
"This is caused by the different implementation "
"that ceil mode in caffe and the floor mode in pytorch.\n"
"You can add the clip layer in caffe prototxt manually "
"if shape mismatch error is caused in caffe. ".format(
layer_name, input.size(), x.size(), caffe_out.size()))
def _hardtanh(raw, input, min_val, max_val, inplace):
# Applies the element-wise function:
# torch.nn.ReLu6
x = raw(input, min_val, max_val)
layer_name = log.add_layer(name='relu6')
log.add_blobs([x], name='relu6_blob')
layer = Layer_param(
name=layer_name,
type='ReLU6',
bottom=[log.blobs(input)],
top=[log.blobs(x)],
)
log.cnet.add_layer(layer)
return x
def _mul(input, *args):
x = raw__mul__(input, *args)
if not NET_INITTED:
return x
layer_name = log.add_layer(name='mul')
top_blobs = log.add_blobs([x], name='mul_blob')
layer = Layer_param(
name=layer_name,
type='Eltwise',
bottom=[log.blobs(input), log.blobs(args[0])],
top=top_blobs,
)
layer.param.eltwise_param.operation = 0 # product is 1
log.cnet.add_layer(layer)
return x
def _linear(raw, input, weight, bias=None):
x = raw(input, weight, bias)
layer_name = log.add_layer(name='fc')
top_blobs = log.add_blobs([x], name='fc_blob')
layer = Layer_param(
name=layer_name,
type='InnerProduct',
bottom=[log.blobs(input)],
top=top_blobs,
)
layer.fc_param(x.size()[1], has_bias=bias is not None)
if bias is not None:
layer.add_data(weight.cpu().data.numpy(), bias.cpu().data.numpy())
else:
layer.add_data(weight.cpu().data.numpy())
log.cnet.add_layer(layer)
return x
def _cat(raw, inputs, dim=0):
x = raw(inputs, dim)
bottom_blobs = []
for input in inputs:
bottom_blobs.append(log.blobs(input))
layer_name = log.add_layer(name='cat')
top_blobs = log.add_blobs([x], name='cat_blob')
layer = Layer_param(
name=layer_name,
type='Concat',
bottom=bottom_blobs,
top=top_blobs,
)
layer.param.concat_param.axis = dim
log.cnet.add_layer(layer)
return x
def _dropout(raw, input, p=0.5, training=False, inplace=False):
x = raw(input, p, training, inplace)
bottom_blobs = [log.blobs(input)]
layer_name = log.add_layer(name='dropout')
top_blobs = log.add_blobs([x], name=bottom_blobs[0], with_num=False)
layer = Layer_param(
name=layer_name,
type='Dropout',
bottom=bottom_blobs,
top=top_blobs,
)
layer.param.dropout_param.dropout_ratio = p
layer.param.include.extend([caffe_pb2.NetStateRule(phase=0)
]) # 1 for test, 0 for train
log.cnet.add_layer(layer)
return x
def _isub(input, *args):
x = raw__isub__(input, *args)
if not NET_INITTED:
return x
x = x.clone()
layer_name = log.add_layer(name='sub')
top_blobs = log.add_blobs([x], name='sub_blob')
layer = Layer_param(
name=layer_name,
type='Eltwise',
bottom=[log.blobs(input), log.blobs(args[0])],
top=top_blobs,
)
layer.param.eltwise_param.operation = 1 # sum is 1
log.cnet.add_layer(layer)
return x
def _softmax(raw, input, dim=None, _stacklevel=3):
# for F.softmax
x = raw(input, dim=dim)
if dim is None:
dim = F._get_softmax_dim('softmax', input.dim(), _stacklevel)
bottom_blobs = [log.blobs(input)]
layer_name = log.add_layer(name='softmax')
log.add_blobs([x], name='softmax_blob')
layer = Layer_param(
name=layer_name,
type='Softmax',
bottom=bottom_blobs,
top=[log.blobs(x)],
)
layer.param.softmax_param.axis = dim
log.cnet.add_layer(layer)
return x
def _add(input, *args):
x = raw__add__(input, *args)
if not NET_INITTED:
return x
layer_name = log.add_layer(name='add')
top_blobs = log.add_blobs([x], name='add_blob')
if log.blobs(args[0]) is None:
log.add_blobs([args[0]], name='extra_blob')
else:
layer = Layer_param(
name=layer_name,
type='Eltwise',
bottom=[log.blobs(input), log.blobs(args[0])],
top=top_blobs,
)
layer.param.eltwise_param.operation = 1 # sum is 1
log.cnet.add_layer(layer)
return x
def _reshape(input, *args):
x = raw_reshape(input, *args)
if not NET_INITTED:
return x
layer_name = log.add_layer(name='reshape')
top_blobs = log.add_blobs([x], name='reshape_blob')
layer = Layer_param(
name=layer_name,
type='Reshape',
bottom=[log.blobs(input)],
top=top_blobs,
)
# TODO: reshpae added to nn_tools layer
dims = list(args)
dims[0] = 0 # the first dim should be batch_size
layer.param.reshape_param.shape.CopyFrom(caffe_pb2.BlobShape(dim=dims))
log.cnet.add_layer(layer)
return x
def _split(raw, input, split_size, dim=0):
# split in pytorch is slice in caffe
x = raw(input, split_size, dim)
layer_name = log.add_layer('split')
top_blobs = log.add_blobs(x, name='split_blob')
layer = Layer_param(
name=layer_name,
type='Slice',
bottom=[log.blobs(input)],
top=top_blobs,
)
slice_num = int(np.floor(input.size()[dim] / split_size))
slice_param = caffe_pb2.SliceParameter(
axis=dim,
slice_point=[split_size * i for i in range(1, slice_num)],
)
layer.param.slice_param.CopyFrom(slice_param)
log.cnet.add_layer(layer)
return x
def _max(raw, *args):
x = raw(*args)
if len(args) == 1:
# TODO max in one tensor
assert NotImplementedError
else:
bottom_blobs = []
for arg in args:
bottom_blobs.append(log.blobs(arg))
layer_name = log.add_layer(name='max')
top_blobs = log.add_blobs([x], name='max_blob')
layer = Layer_param(
name=layer_name,
type='Eltwise',
bottom=bottom_blobs,
top=top_blobs,
)
layer.param.eltwise_param.operation = 2
log.cnet.add_layer(layer)
return x
def _mean(input, *args, **kwargs):
x = raw_mean(input, *args, **kwargs)
if not NET_INITTED:
return x
layer_name = log.add_layer(name='mean')
top_blobs = log.add_blobs([x], name='mean_blob')
layer = Layer_param(
name=layer_name,
type='Reduction',
bottom=[log.blobs(input)],
top=top_blobs,
)
if len(args) == 1:
dim = args[0]
elif 'dim' in kwargs:
dim = kwargs['dim']
else:
raise NotImplementedError('mean operation must specify a dim')
layer.param.reduction_param.operation = 4
layer.param.reduction_param.axis = dim
log.cnet.add_layer(layer)
return x
def _l2Norm(raw, input, weight, eps):
# Applies the element-wise function:
# L2Norm in vgg_ssd
x = raw(input, weight, eps)
layer_name = log.add_layer(name='normalize')
log.add_blobs([x], name='normalize_blob')
layer = Layer_param(
name=layer_name,
type='Normalize',
bottom=[log.blobs(input)],
top=[log.blobs(x)],
)
layer.norm_param(eps)
layer.add_data(weight.cpu().data.numpy())
log.cnet.add_layer(layer)
return x
def _prelu(raw, input, weight):
# for threshold or prelu
x = raw(input, weight)
bottom_blobs = [log.blobs(input)]
layer_name = log.add_layer(name='prelu')
log.add_blobs([x], name='prelu_blob')
layer = Layer_param(
name=layer_name,
type='PReLU',
bottom=bottom_blobs,
top=[log.blobs(x)],
)
if weight.size()[0] == 1:
layer.param.prelu_param.channel_shared = True
layer.add_data(weight.cpu().data.numpy()[0])
else:
layer.add_data(weight.cpu().data.numpy())
log.cnet.add_layer(layer)
return x
def _batch_norm(
raw,
input,
running_mean,
running_var,
weight=None,
bias=None,
training=False,
momentum=0.1,
eps=1e-5,
):
# because the runing_mean and runing_var will be changed after
# the _batch_norm operation, we first save the parameters
x = raw(input, running_mean, running_var, weight, bias, training, momentum,
eps)
bottom_blobs = [log.blobs(input)]
layer_name1 = log.add_layer(name='batch_norm')
top_blobs = log.add_blobs([x], name='batch_norm_blob')
layer1 = Layer_param(
name=layer_name1,
type='BatchNorm',
bottom=bottom_blobs,
top=top_blobs,
)
if running_mean is None or running_var is None:
# not use global_stats, normalization is performed over the current mini-batch
layer1.batch_norm_param(use_global_stats=0, eps=eps)
else:
layer1.batch_norm_param(use_global_stats=1, eps=eps)
running_mean_clone = running_mean.clone()
running_var_clone = running_var.clone()
layer1.add_data(
running_mean_clone.cpu().numpy(),
running_var_clone.cpu().numpy(),
np.array([1.0]),
)
log.cnet.add_layer(layer1)
if weight is not None and bias is not None:
layer_name2 = log.add_layer(name='bn_scale')
layer2 = Layer_param(
name=layer_name2,
type='Scale',
bottom=top_blobs,
top=top_blobs,
)
layer2.param.scale_param.bias_term = True
layer2.add_data(weight.cpu().data.numpy(), bias.cpu().data.numpy())
log.cnet.add_layer(layer2)
return x
def _instance_norm(
raw,
input,
running_mean=None,
running_var=None,
weight=None,
bias=None,
use_input_stats=True,
momentum=0.1,
eps=1e-5,
):
# TODO: the batch size!=1 view operations
print(
"WARNING: The Instance Normalization transfers to Caffe "
"using BatchNorm, so the batch size should be 1", )
if running_var is not None or weight is not None:
# TODO: the affine=True or track_running_stats=True case
raise NotImplementedError(
"not implement the affine=True or track_running_stats=True case InstanceNorm"
)
x = torch.batch_norm(
input,
weight,
bias,
running_mean,
running_var,
use_input_stats,
momentum,
eps,
torch.backends.cudnn.enabled,
)
bottom_blobs = [log.blobs(input)]
layer_name1 = log.add_layer(name='instance_norm')
top_blobs = log.add_blobs([x], name='instance_norm_blob')
layer1 = Layer_param(
name=layer_name1,
type='BatchNorm',
bottom=bottom_blobs,
top=top_blobs,
)
if running_mean is None or running_var is None:
# not use global_stats, normalization is performed over the current mini-batch
layer1.batch_norm_param(use_global_stats=0, eps=eps)
running_mean = torch.zeros(input.size()[1])
running_var = torch.ones(input.size()[1])
else:
layer1.batch_norm_param(use_global_stats=1, eps=eps)
running_mean_clone = running_mean.clone()
running_var_clone = running_var.clone()
layer1.add_data(
running_mean_clone.cpu().numpy(),
running_var_clone.cpu().numpy(),
np.array([1.0]),
)
log.cnet.add_layer(layer1)
if weight is not None and bias is not None:
layer_name2 = log.add_layer(name='bn_scale')
layer2 = Layer_param(
name=layer_name2,
type='Scale',
bottom=top_blobs,
top=top_blobs,
)
layer2.param.scale_param.bias_term = True
layer2.add_data(
weight.cpu().data.numpy(),
bias.cpu().data.numpy(),
)
log.cnet.add_layer(layer2)
return x