InceptionBlock

class eyefeatures.deep.models.InceptionBlock(in_channels, ch1x1, ch3x3_reduce, ch3x3, ch5x5_reduce, ch5x5, pool_proj)[source]

Bases: Module

An Inception-style block consisting of multiple convolution branches operating on different scales.

Parameters:

  • in_channels (int) – (int) Number of input channels.

  • ch1x1 (int) – (int) Number of output channels for the 1x1 convolution branch.

  • ch3x3_reduce (int) – (int) Number of output channels for the 1x1 convolution before the 3x3 convolution.

  • ch3x3 (int) – (int) Number of output channels for the 3x3 convolution branch.

  • ch5x5_reduce (int) – (int) Number of output channels for the 1x1 convolution before the 5x5 convolution.

  • ch5x5 (int) – (int) Number of output channels for the 5x5 convolution branch.

  • pool_proj (int) – (int) Number of output channels for the 1x1 convolution after the max pooling branch.

Returns:

Concatenated output tensor from all branches.