What we are doing through 1X1 Convolution Layer is we are just reducing the number of layers, that we can do in 3X3 layers also
Why 1x1 conv Layer in GoogLe Net?
Hey,
The concept of inception module is bit different, It combines filter maps from different different operations into one, so that we get a lot of filter maps (channels)
If we would do only 1 convolution operation with 3x3xC it will create C channels in activation maps. But in inception module we are getting (C1 + C2 + C3 + C3) these channels in output, more number of channels means more features are captured. therefore it is quite an efficient way to combine multiple operations into one.
Thanks 