image implementation

[hhhhq12]

For the RGB image implementation, does the network utilize weight sharing across the three channels, or are the R, G, and B channels treated as a concatenated input processed by a single network?