private _inceptionblock — function
_inceptionblock(inplanes, out_1x1, red_3x3, out_3x3, red_5x5, out_3x3, pool_proj)
Create an inception module for use in GoogLeNet (reference).
Arguments
- inplanes: the number of input feature maps
- out_1x1: the number of output feature maps for the 1x1 convolution (branch 1)
- red_3x3: the number of output feature maps for the 3x3 reduction convolution (branch 2)
- out_3x3: the number of output feature maps for the 3x3 convolution (branch 2)
- red_5x5: the number of output feature maps for the 5x5 reduction convolution (branch 3)
- out_5x5: the number of output feature maps for the 5x5 convolution (branch 3)
- pool_proj: the number of output feature maps for the pooling projection (branch 4)