Conv

class tripy.Conv(in_channels: int, out_channels: int, kernel_dims: Sequence[int], padding: Sequence[Sequence[int]] | None = None, stride: Sequence[int] | None = None, groups: int | None = None, dilation: Sequence[int] | None = None, bias: bool = True, dtype: dtype = float32)[source]

Applies a convolution on the input tensor.

With an input of shape \((N, C_{\text{in}}, D_0,\ldots,D_n)\) and output of shape \((N, C_{\text{out}}, D_{0_{\text{out}}},\ldots,D_{n_{\text{out}}})\) the output values are given by:

\[\text{out}(N_i, C_{\text{out}_j}) = \text{Bias}_{C_{\text{out}}} + \sum_{k = 0}^{C_{\text{in}} - 1} \text{weight}(C_{\text{out}_j}, k) \star \text{input}(N_i, k)\]

where \(\star\) is the cross-correlation operator applied over the spatial dimensions of the input and kernel, \(N\) is the batch dimension, \(C\) is the channel dimension, and \(D_0,\ldots,D_n\) are the spatial dimensions.

Parameters:

  • in_channels (int) – The number of channels in the input tensor.

  • out_channels (int) – The number of channels produced by the convolution.

  • kernel_dims (Sequence[int]) – The spatial shape of the kernel.

  • padding (Sequence[Sequence[int]]) – A sequence of pairs of integers of length \(M\) indicating the zero padding to apply to the input along each spatial dimension before and after the dimension respectively, where \(M\) is the number of spatial dimensions, i.e. \(M = \text{rank(input)} - 2\). Defaults to all 0.

  • stride (Sequence[int]) – A sequence of length \(M\) indicating the stride of convolution across each spatial dimension, where \(M\) is the number of spatial dimensions, i.e. \(M = \text{rank(input)} - 2\). Defaults to all 1.

  • groups (int) – The number of groups in a grouped convolution where the input and output channels are divided into groups groups. Each output group is connected only to its corresponding input group through the convolution kernel weights, and the outputs for each group are concatenated to produce the final result. This is in contrast to a standard convolution which has full connectivity between all input and output channels. Grouped convolutions reduce computational cost by a factor of groups and can benefit model parallelism and memory usage. Note that in_channels and out_channels must both be divisible by groups. Defaults to 1 (standard convolution).

  • dilation (Sequence[int]) – A sequence of length \(M\) indicating the number of zeros to insert between kernel weights across each spatial dimension, where \(M\) is the number of spatial dimensions, i.e. \(M = \text{rank(input)} - 2\). This is known as the a trous algorithm and further downsamples the output by increasing the receptive field of the kernel. For each dimension with value \(x\), \(x-1\) zeros are inserted between kernel weights.

  • bias (Parameter | None) – Whether to add a bias term to the output or not. The bias has a shape of \((\text{out_channels},)\).

  • dtype (dtype) – The data type to use for the convolution weights.

Example
Example
1input = tp.reshape(tp.arange(16, dtype=tp.float32), (1, 1, 4, 4))
2conv = tp.Conv(in_channels=1, out_channels=1, kernel_dims=(2, 2), dtype=tp.float32)
3output = conv(input)

>>> input
tensor(
    [[[[0.0000, 1.0000, 2.0000, 3.0000],
       [4.0000, 5.0000, 6.0000, 7.0000],
       [8.0000, 9.0000, 10.0000, 11.0000],
       [12.0000, 13.0000, 14.0000, 15.0000]]]], 
    dtype=float32, loc=gpu:0, shape=(1, 1, 4, 4))
>>> conv.state_dict()
{
    bias: tensor([0.0000], dtype=float32, loc=gpu:0, shape=(1,)),
    weight: tensor(
        [[[[0.0000, 1.0000],
           [2.0000, 3.0000]]]], 
        dtype=float32, loc=gpu:0, shape=(1, 1, 2, 2)),
}
>>> output
tensor(
    [[[[24.0000, 30.0000, 36.0000],
       [48.0000, 54.0000, 60.0000],
       [72.0000, 78.0000, 84.0000]]]], 
    dtype=float32, loc=gpu:0, shape=(1, 1, 3, 3))
Example: Using Padding and Stride
Using Padding and Stride
1input = tp.reshape(tp.arange(16, dtype=tp.float32), (1, 1, 4, 4))
2conv = tp.Conv(1, 1, (3, 3), padding=((1, 1), (1, 1)), stride=(3, 1), bias=False, dtype=tp.float32)
3output = conv(input)

>>> input
tensor(
    [[[[0.0000, 1.0000, 2.0000, 3.0000],
       [4.0000, 5.0000, 6.0000, 7.0000],
       [8.0000, 9.0000, 10.0000, 11.0000],
       [12.0000, 13.0000, 14.0000, 15.0000]]]], 
    dtype=float32, loc=gpu:0, shape=(1, 1, 4, 4))
>>> conv.state_dict()
{
    weight: tensor(
        [[[[0.0000, 1.0000, 2.0000],
           [3.0000, 4.0000, 5.0000],
           [6.0000, 7.0000, 8.0000]]]], 
        dtype=float32, loc=gpu:0, shape=(1, 1, 3, 3)),
}
>>> output
tensor(
    [[[[73.0000, 121.0000, 154.0000, 103.0000],
       [139.0000, 187.0000, 202.0000, 113.0000]]]], 
    dtype=float32, loc=gpu:0, shape=(1, 1, 2, 4))
Example: Depthwise Convolution
Depthwise Convolution
1input = tp.reshape(tp.arange(18, dtype=tp.float32), (1, 2, 3, 3))
2conv = tp.Conv(2, 2, (3, 3), groups=2, bias=False, dtype=tp.float32)
3output = conv(input)

>>> input
tensor(
    [[[[0.0000, 1.0000, 2.0000],
       [3.0000, 4.0000, 5.0000],
       [6.0000, 7.0000, 8.0000]],

      [[9.0000, 10.0000, 11.0000],
       [12.0000, 13.0000, 14.0000],
       [15.0000, 16.0000, 17.0000]]]], 
    dtype=float32, loc=gpu:0, shape=(1, 2, 3, 3))
>>> conv.state_dict()
{
    weight: tensor(
        [[[[0.0000, 1.0000, 2.0000],
           [3.0000, 4.0000, 5.0000],
           [6.0000, 7.0000, 8.0000]]],


         [[[9.0000, 10.0000, 11.0000],
           [12.0000, 13.0000, 14.0000],
           [15.0000, 16.0000, 17.0000]]]], 
        dtype=float32, loc=gpu:0, shape=(2, 1, 3, 3)),
}
>>> output
tensor(
    [[[[204.0000]],

      [[1581.0000]]]], 
    dtype=float32, loc=gpu:0, shape=(1, 2, 1, 1))
Example: Dilated Convolution (a trous algorithm)
Dilated Convolution (a trous algorithm)
1input = tp.reshape(tp.arange(9, dtype=tp.float32), (1, 1, 3, 3))
2conv = tp.Conv(1, 1, (2, 2), dilation=(2, 2), bias=False, dtype=tp.float32)
3output = conv(input)

>>> input
tensor(
    [[[[0.0000, 1.0000, 2.0000],
       [3.0000, 4.0000, 5.0000],
       [6.0000, 7.0000, 8.0000]]]], 
    dtype=float32, loc=gpu:0, shape=(1, 1, 3, 3))
>>> conv.state_dict()
{
    weight: tensor(
        [[[[0.0000, 1.0000],
           [2.0000, 3.0000]]]], 
        dtype=float32, loc=gpu:0, shape=(1, 1, 2, 2)),
}
>>> output
tensor([[[[38.0000]]]], dtype=float32, loc=gpu:0, shape=(1, 1, 1, 1))
dtype: dtype

The data type to use for the convolution weights.

padding: Sequence[Sequence[int]]

A sequence of pairs of integers of length \(M\) indicating the zero padding to apply to the input along each spatial dimension before and after the dimension respectively, where \(M\) is the number of spatial dimensions, i.e. \(M = \text{rank(input)} - 2\).

stride: Sequence[int]

A sequence of length \(M\) indicating the stride of convolution across each spatial dimension, where \(M\) is the number of spatial dimensions, i.e. \(M = \text{rank(input)} - 2\).

groups: int

The number of groups in a grouped convolution where the input and output channels are divided into groups groups. Each output group is connected only to its corresponding input group through the convolution kernel weights, and the outputs for each group are concatenated to produce the final result. This is in contrast to a standard convolution which has full connectivity between all input and output channels. Grouped convolutions reduce computational cost by a factor of groups and can benefit model parallelism and memory usage. Note that in_channels and out_channels must both be divisible by groups.

dilation: Sequence[int]

A sequence of length \(M\) indicating the number of zeros to insert between kernel weights across each spatial dimension, where \(M\) is the number of spatial dimensions, i.e. \(M = \text{rank(input)} - 2\). This is known as the a trous algorithm and further downsamples the output by increasing the receptive field of the kernel. For each dimension with value \(x\), \(x-1\) zeros are inserted between kernel weights.

bias: Parameter | None

The bias term to add to the output. The bias has a shape of \((\text{out_channels},)\).

weight: Parameter

The kernel of shape \((\text{out_channels}, \frac{\text{in_channels}}{\text{groups}}, *\text{kernel_dims})\).

__call__(input: Tensor) → Tensor[source]
Parameters:

input (Tensor) – The input tensor.

Returns:

A tensor of the same data type as the input with a shape \((N, \text{out_channels}, D_{0_{\text{out}}},\ldots,D_{n_{\text{out}}})\) where \(D_{k_{\text{out}}} = \large \left\lfloor \frac{D_{k_{\text{in}}} + \text{padding}_{k_0} + \text{padding}_{k_1} - \text{dilation}_k \times (\text{kernel_dims}_k - 1) - 1}{\text{stride}_k} \right\rfloor + \normalsize 1\)

Return type:

Tensor