Conv

Function

Conv2d: extracts local spatial features from images or feature maps.

Conv3d: performs convolution operations on 3D volume data to extract spatiotemporal features.

Prototype

torch.nn.Conv2d(
    in_channels: int,
    out_channels: int,
    kernel_size: Union[int, Tuple[int, int]],
    stride: Union[int, Tuple[int, int]] = 1,
    padding: Union[str, int, Tuple[int, int]] = 0,
    dilation: Union[int, Tuple[int, int]] = 1,
    groups: int = 1,
    bias: bool = True,
    padding_mode: str = 'zeros',
    device=None,
    dtype=None
)
torch.nn.Conv3d(
    in_channels: int,
    out_channels: int,
    kernel_size: Union[int, Tuple[int, int, int]],
    stride: Union[int, Tuple[int, int, int]] = 1,
    padding: Union[str, int, Tuple[int, int, int]] = 0,
    dilation: Union[int, Tuple[int, int, int]] = 1,
    groups: int = 1,
    bias: bool = True,
    padding_mode: str = 'zeros',
    device=None,
    dtype=None
)

Parameters

**Table 1** Conv2d parameters
Parameter	Type	Mandatory (Yes/No)	Description
in_channels	int	Yes	Number of input channels
out_channels	int	Yes	Number of output channels (number of convolution kernels)
kernel_size	int/tuple	Yes	Convolution kernel size (for example, 3 or (3, 3))
stride	int/tuple	No	Stride (default value: 1)
padding	int/tuple/str	No	Border padding (default value: 0; other values include same, etc.)
dilation	int/tuple	No	Dilated convolution rate (default value: 1)
groups	int	No	Number of groups for grouped convolution (default value: 1; in_channels must be exactly divided by the number of groups)
bias	bool	No	Whether to add a bias (default value: True)
padding_mode	str	No	Padding mode, including zeros, reflect, replicate, and circular

**Table 2** Conv3d parameters
Parameter	Type	Mandatory (Yes/No)	Description
in_channels	int	Yes	Number of channels of the input volume data (for example, 3 for RGB videos)
out_channels	int	Yes	Number of output channels (number of convolution kernels)
kernel_size	int/tuple	Yes	Size of the 3D convolution kernel (for example, (3, 3, 3) or 5)
stride	int/tuple	No	3D stride (default value: (1, 1, 1))
padding	int/tuple/str	No	3D border padding (default value: 0; other values include same, etc.)
dilation	int/tuple	No	Dilated 3D convolution rate (default value: 1)
groups	int	No	Number of groups for grouped convolution (default value: 1)
bias	bool	No	Whether to add a bias (default value: True)
padding_mode	str	No	Padding mode, including zeros, reflect, replicate, and circular

Parent topic: Operator Interfaces