这里写自定义目录标题
CNN卷积神经网络理解参数尺寸计算理解
图片尺寸大小经过卷积后计算
卷积尺寸计算:尺寸变为((原始尺寸+2*填充-核尺寸)/步长)+1
Conv2d参数如下:
conv2d 参数(输入通道,输出通道,核尺寸,步长,填充等)
in_channels: int,
out_channels: int,
kernel_size: _size_2_t,
stride: _size_2_t = 1,
padding: Union[str, _size_2_t] = 0,
dilation: _size_2_t = 1,
groups: int = 1,
bias: bool = True,
padding_mode: str = 'zeros', # TODO: refine this type
device=None,
dtype=None
卷积尺寸计算:经过Conv2d卷积后,尺寸变为((原始尺寸+2填充-核尺寸)/步长)+1
举个例子,卷积自编码为例,原始数据为1通道,2828的图片
class AutoEncoder(nn.Module):
def __init__(self):
super(AutoEncoder, self).__init__()
self.en_conv = nn.Sequential(
nn.Conv2d(1, 16, 4, 2, 1), 原始图像 1通道 尺寸 28*28,核尺寸4*4,步长为2,填充1
nn.BatchNorm2d(16), 经过一次卷积 16通道 14*14 按上面公式计算的 (28+2-4)/2 +1 =14
nn.Tanh(),
nn.Conv2d(16, 32, 4, 2, 1),
nn.BatchNorm2d(32), 经过一次卷积 32通道 7*7
nn.Tanh(),
nn.Conv2d(32, 16, 3, 1, 1),
nn.BatchNorm2d(16), 经过一次卷积 16通道 7*7
nn.Tanh()
)
self.en_fc = nn.Linear(16*7*7, HIDDEN_SIZE)
self.de_fc = nn.Linear(HIDDEN_SIZE, 16*7*7)
self.de_conv = nn.Sequential(
nn.ConvTranspose2d(16, 16, 4, 2, 1), 反卷积,可以参考反卷积计算公式
nn.BatchNorm2d(16),
nn.Tanh(),
nn.ConvTranspose2d(16, 1, 4, 2, 1),
nn.Sigmoid()
)
def forward(self, x):
en = self.en_conv(x)
code = self.en_fc(en.view(en.size(0), -1))
de = self.de_fc(code)
decoded = self.de_conv(de.view(de.size(0), 16, 7, 7))
return code, decoded`
``