GoogLeNet-FCN architecture

[mrpaulbotezatu]

Could you please help me with some queries that I have regarding the architecture of FCN GoogLe Net that you presented it? I have tried to implement myself a GoogLeNet-FCN architecture but it doesn't work.

The blue lines in the model are the upsampling paths that need to be realised? For example the blue drawn line which is under the FCN8 word reffers to this upsampling path? : "Sums the 2x upsampled conv7 (with a stride 2 transposed convolution) with pool4, upsamples them with a stride 2 transposed convolution and sums them with pool3, and applies a transposed convolution layer with stride 8 on the resulting feature maps to obtain the segmentation map."