Conv3d#
- class Conv3d(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True, conv_mode='cross_correlation')[源代码]#
对输入 tensor 进行三维卷积
例如,给一个大小为 \((N, C_{\text{in}}, T, H, W)`的输入, 该层会通过下述过程生成大小为 :math:`(N, C_{\text{out}}, T_{\text{out}}, H_{\text{out}}, W_{\text{out}})\) 的输出:
\[\text{out}(N_i, C_{\text{out}_j}) = \text{bias}(C_{\text{out}_j}) + \sum_{k = 0}^{C_{\text{in}} - 1} \text{weight}(C_{\text{out}_j}, k) \star \text{input}(N_i, k)\]在此式子中 \(\star\) 是有效的 3D 互相关(cross-correlation) 运算符, \(N\) 是 batch 大小, \(C\) 表示 channels 数量。
当 groups == in_channels 且 out_channels == K * in_channels ,其中 K 是正整数,该操作也被称为深度方向卷积(depthwise convolution)。
In other words, for an input of size \((N, C_{in}, T_{int}, H_{in}, W_{in})\), a depthwise convolution with a depthwise multiplier K, can be constructed by arguments \((in\_channels=C_{in}, out\_channels=C_{in} \times K, ..., groups=C_{in})\).
- 参数:
in_channels (
int
) – 输入数据中的通道数。out_channels (
int
) – 输出数据中的通道数。kernel_size (
Union
[int
,Tuple
[int
,int
,int
]]) – 空间维度上的权重大小。如果kernel_size 是一个int
, 实际的kernel大小为 (kernel_size, kernel_size, kernel_size)。stride (
Union
[int
,Tuple
[int
,int
,int
]]) – stride of the 3D convolution operation. Default: 1padding (
Union
[int
,Tuple
[int
,int
,int
]]) – size of the paddings added to the input on both sides of its spatial dimensions. Only zero-padding is supported. Default: 0dilation (
Union
[int
,Tuple
[int
,int
,int
]]) – dilation of the 3D convolution operation. Default: 1groups (
int
) – number of groups into which the input and output channels are divided, so as to perform agrouped convolution
. Whengroups
is not 1,in_channels
andout_channels
must be divisible bygroups
, and the shape of weight should be(groups, out_channel // groups, in_channels // groups, depth, height, width)
. Default: 1bias (
bool
) – whether to add a bias onto the result of convolution. Default: Trueconv_mode (
str
) – Supports cross_correlation. Default: cross_correlation
备注
weight
的shape通常是(out_channels, in_channels, depth, height, width)
, 如果 groups 不为1, shape 将是(groups, out_channels // groups, in_channels // groups, depth, height, width)
bias
的shape通常是(1, out_channels, *1)
实际案例
>>> import numpy as np >>> m = M.Conv3d(in_channels=3, out_channels=1, kernel_size=3) >>> inp = mge.tensor(np.arange(0, 384).astype("float32").reshape(2, 3, 4, 4, 4)) >>> oup = m(inp) >>> oup.numpy().shape (2, 1, 2, 2, 2)