"""
MindSpore implementation of `InceptionV3`.
Refer to Rethinking the Inception Architecture for Computer Vision.
"""
from typing import Union, Tuple
from mindspore import nn, ops, Tensor
import mindspore.common.initializer as init
from .utils import load_pretrained
from .registry import register_model
from .layers.pooling import GlobalAvgPooling
__all__ = [
'InceptionV3',
'inception_v3'
]
def _cfg(url='', **kwargs):
return {
'url': url,
'num_classes': 1000,
'first_conv': 'conv1a', 'classifier': 'classifier',
**kwargs
}
default_cfgs = {
'inception_v3': _cfg(url='')
}
class BasicConv2d(nn.Cell):
"""A block for conv bn and relu"""
def __init__(self,
in_channels: int,
out_channels: int,
kernel_size: Union[int, Tuple] = 1,
stride: int = 1,
padding: int = 0,
pad_mode: str = 'same'
) -> None:
super().__init__()
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size, stride,
padding=padding, pad_mode=pad_mode)
self.bn = nn.BatchNorm2d(out_channels, eps=0.001, momentum=0.9997)
self.relu = nn.ReLU()
def construct(self, x: Tensor) -> Tensor:
x = self.conv(x)
x = self.bn(x)
x = self.relu(x)
return x
class InceptionA(nn.Cell):
def __init__(self,
in_channels: int,
pool_features: int
) -> None:
super().__init__()
self.branch0 = BasicConv2d(in_channels, 64, kernel_size=1)
self.branch1 = nn.SequentialCell([
BasicConv2d(in_channels, 48, kernel_size=1),
BasicConv2d(48, 64, kernel_size=5)
])
self.branch2 = nn.SequentialCell([
BasicConv2d(in_channels, 64, kernel_size=1),
BasicConv2d(64, 96, kernel_size=3),
BasicConv2d(96, 96, kernel_size=3)
])
self.branch_pool = nn.SequentialCell([
nn.AvgPool2d(kernel_size=3, pad_mode='same'),
BasicConv2d(in_channels, pool_features, kernel_size=1)
])
def construct(self, x: Tensor) -> Tensor:
x0 = self.branch0(x)
x1 = self.branch1(x)
x2 = self.branch2(x)
branch_pool = self.branch_pool(x)
out = ops.concat((x0, x1, x2, branch_pool), axis=1)
return out
class InceptionB(nn.Cell):
def __init__(self, in_channels: int) -> None:
super().__init__()
self.branch0 = BasicConv2d(in_channels, 384, kernel_size=3, stride=2, pad_mode='valid')
self.branch1 = nn.SequentialCell([
BasicConv2d(in_channels, 64, kernel_size=1),
BasicConv2d(64, 96, kernel_size=3),
BasicConv2d(96, 96, kernel_size=3, stride=2, pad_mode='valid')
])
self.branch_pool = nn.MaxPool2d(kernel_size=3, stride=2)
def construct(self, x: Tensor) -> Tensor:
x0 = self.branch0(x)
x1 = self.branch1(x)
branch_pool = self.branch_pool(x)
out = ops.concat((x0, x1, branch_pool), axis=1)
return out
class InceptionC(nn.Cell):
def __init__(self,
in_channels: int,
channels_7x7: int
) -> None:
super().__init__()
self.branch0 = BasicConv2d(in_channels, 192, kernel_size=1)
self.branch1 = nn.SequentialCell([
BasicConv2d(in_channels, channels_7x7, kernel_size=1),
BasicConv2d(channels_7x7, channels_7x7, kernel_size=(1, 7)),
BasicConv2d(channels_7x7, 192, kernel_size=(7, 1))
])
self.branch2 = nn.SequentialCell([
BasicConv2d(in_channels, channels_7x7, kernel_size=1),
BasicConv2d(channels_7x7, channels_7x7, kernel_size=(7, 1)),
BasicConv2d(channels_7x7, channels_7x7, kernel_size=(1, 7)),
BasicConv2d(channels_7x7, channels_7x7, kernel_size=(7, 1)),
BasicConv2d(channels_7x7, 192, kernel_size=(1, 7))
])
self.branch_pool = nn.SequentialCell([
nn.AvgPool2d(kernel_size=3, pad_mode='same'),
BasicConv2d(in_channels, 192, kernel_size=1)
])
def construct(self, x: Tensor) -> Tensor:
x0 = self.branch0(x)
x1 = self.branch1(x)
x2 = self.branch2(x)
branch_pool = self.branch_pool(x)
out = ops.concat((x0, x1, x2, branch_pool), axis=1)
return out
class InceptionD(nn.Cell):
def __init__(self, in_channels: int) -> None:
super().__init__()
self.branch0 = nn.SequentialCell([
BasicConv2d(in_channels, 192, kernel_size=1),
BasicConv2d(192, 320, kernel_size=3, stride=2, pad_mode='valid')
])
self.branch1 = nn.SequentialCell([
BasicConv2d(in_channels, 192, kernel_size=1),
BasicConv2d(192, 192, kernel_size=(1, 7)), # check
BasicConv2d(192, 192, kernel_size=(7, 1)),
BasicConv2d(192, 192, kernel_size=3, stride=2, pad_mode='valid')
])
self.branch_pool = nn.MaxPool2d(kernel_size=3, stride=2)
def construct(self, x: Tensor) -> Tensor:
x0 = self.branch0(x)
x1 = self.branch1(x)
branch_pool = self.branch_pool(x)
out = ops.concat((x0, x1, branch_pool), axis=1)
return out
class InceptionE(nn.Cell):
def __init__(self, in_channels: int) -> None:
super().__init__()
self.branch0 = BasicConv2d(in_channels, 320, kernel_size=1)
self.branch1 = BasicConv2d(in_channels, 384, kernel_size=1)
self.branch1a = BasicConv2d(384, 384, kernel_size=(1, 3))
self.branch1b = BasicConv2d(384, 384, kernel_size=(3, 1))
self.branch2 = nn.SequentialCell([
BasicConv2d(in_channels, 448, kernel_size=1),
BasicConv2d(448, 384, kernel_size=3)
])
self.branch2a = BasicConv2d(384, 384, kernel_size=(1, 3))
self.branch2b = BasicConv2d(384, 384, kernel_size=(3, 1))
self.branch_pool = nn.SequentialCell([
nn.AvgPool2d(kernel_size=3, pad_mode='same'),
BasicConv2d(in_channels, 192, kernel_size=1)
])
def construct(self, x: Tensor) -> Tensor:
x0 = self.branch0(x)
x1 = self.branch1(x)
x1 = ops.concat((self.branch1a(x1), self.branch1b(x1)), axis=1)
x2 = self.branch2(x)
x2 = ops.concat((self.branch2a(x2), self.branch2b(x2)), axis=1)
branch_pool = self.branch_pool(x)
out = ops.concat((x0, x1, x2, branch_pool), axis=1)
return out
class InceptionAux(nn.Cell):
"""Inception module for the aux classifier head"""
def __init__(self,
in_channels: int,
num_classes: int
) -> None:
super().__init__()
self.avg_pool = nn.AvgPool2d(5, stride=3, pad_mode='valid')
self.conv0 = BasicConv2d(in_channels, 128, kernel_size=1)
self.conv1 = BasicConv2d(128, 768, kernel_size=5, pad_mode='valid')
self.flatten = nn.Flatten()
self.fc = nn.Dense(in_channels, num_classes)
def construct(self, x: Tensor) -> Tensor:
x = self.avg_pool(x)
x = self.conv0(x)
x = self.conv1(x)
x = self.flatten(x)
x = self.fc(x)
return x
[文档]class InceptionV3(nn.Cell):
r"""Inception v3 model architecture from
`"Rethinking the Inception Architecture for Computer Vision" <https://arxiv.org/abs/1512.00567>`_.
.. note::
**Important**: In contrast to the other models the inception_v3 expects tensors with a size of
N x 3 x 299 x 299, so ensure your images are sized accordingly.
Args:
num_classes: number of classification classes. Default: 1000.
aux_logits: use auxiliary classifier or not. Default: False.
in_channels: number the channels of the input. Default: 3.
drop_rate: dropout rate of the layer before main classifier. Default: 0.2.
"""
def __init__(self,
num_classes: int = 1000,
aux_logits: bool = True,
in_channels: int = 3,
drop_rate: float = 0.2) -> None:
super().__init__()
self.aux_logits = aux_logits
self.conv1a = BasicConv2d(in_channels, 32, kernel_size=3, stride=2, pad_mode='valid')
self.conv2a = BasicConv2d(32, 32, kernel_size=3, stride=1, pad_mode='valid')
self.conv2b = BasicConv2d(32, 64, kernel_size=3, stride=1)
self.maxpool1 = nn.MaxPool2d(kernel_size=3, stride=2)
self.conv3b = BasicConv2d(64, 80, kernel_size=1)
self.conv4a = BasicConv2d(80, 192, kernel_size=3, pad_mode='valid')
self.maxpool2 = nn.MaxPool2d(kernel_size=3, stride=2)
self.inception5b = InceptionA(192, pool_features=32)
self.inception5c = InceptionA(256, pool_features=64)
self.inception5d = InceptionA(288, pool_features=64)
self.inception6a = InceptionB(288)
self.inception6b = InceptionC(768, channels_7x7=128)
self.inception6c = InceptionC(768, channels_7x7=160)
self.inception6d = InceptionC(768, channels_7x7=160)
self.inception6e = InceptionC(768, channels_7x7=192)
if self.aux_logits:
self.aux = InceptionAux(768, num_classes)
self.inception7a = InceptionD(768)
self.inception7b = InceptionE(1280)
self.inception7c = InceptionE(2048)
self.pool = GlobalAvgPooling()
self.dropout = nn.Dropout(keep_prob=1 - drop_rate)
self.num_features = 2048
self.classifier = nn.Dense(self.num_features, num_classes)
self._initialize_weights()
def _initialize_weights(self) -> None:
"""Initialize weights for cells."""
for _, cell in self.cells_and_names():
if isinstance(cell, nn.Conv2d):
cell.weight.set_data(
init.initializer(init.XavierUniform(), cell.weight.shape, cell.weight.dtype))
def forward_preaux(self, x: Tensor) -> Tensor:
x = self.conv1a(x)
x = self.conv2a(x)
x = self.conv2b(x)
x = self.maxpool1(x)
x = self.conv3b(x)
x = self.conv4a(x)
x = self.maxpool2(x)
x = self.inception5b(x)
x = self.inception5c(x)
x = self.inception5d(x)
x = self.inception6a(x)
x = self.inception6b(x)
x = self.inception6c(x)
x = self.inception6d(x)
x = self.inception6e(x)
return x
def forward_postaux(self, x: Tensor) -> Tensor:
x = self.inception7a(x)
x = self.inception7b(x)
x = self.inception7c(x)
return x
def forward_features(self, x: Tensor) -> Tensor:
x = self.forward_preaux(x)
x = self.forward_postaux(x)
return x
def construct(self, x: Tensor) -> Union[Tensor, Tuple[Tensor, Tensor]]:
x = self.forward_preaux(x)
if self.training and self.aux_logits:
aux = self.aux(x)
else:
aux = None
x = self.forward_postaux(x)
x = self.pool(x)
x = self.dropout(x)
x = self.classifier(x)
if self.training and self.aux_logits:
return x, aux
return x
@register_model
def inception_v3(pretrained: bool = False, num_classes: int = 1000, in_channels=3, **kwargs) -> InceptionV3:
"""Get InceptionV3 model.
Refer to the base class `models.InceptionV3` for more details."""
default_cfg = default_cfgs['inception_v3']
model = InceptionV3(num_classes=num_classes, aux_logits=True, in_channels=in_channels, **kwargs)
if pretrained:
load_pretrained(model, default_cfg, num_classes=num_classes, in_channels=in_channels)
return model