Comment by frumiousirc
2 days ago
After reading the paper there is one thing I don't understand about the DDL. It seems each "concat" will increase the size of the "output feature" relative to the "input feature" by the size of the "generated image".
Is that right?
If so, how is this increased size handled by each downstream DDL?
Or, is there a 2x pooling in the "concat" step so that final size remains unchanged?
Yes, there is a transform that make final size of stem features remains unchanged