Hi I am trying to design a GCN encoder - decoder network. My input shape was 330025 and after some GCN layers followed by average pooling the encoded shape is 25611. I want my decoder to produce the input shape, so that I can compare the original and reproduced data. But I am lost how can I do it.
My Encoder:
input = [N, 3, 300, 25]
h8 = [N, 256, 75, 25]
encode = F.avg_pool2d(h8, h8.size()[2:]) #shape [N, 256, 1, 1]
How can I use this “encode” to increase to original shape.
I am using AS-GCN blocks based on ST-GCN from this repo