This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Figure 1.
Structure diagram of CSRGAN.
Figure 2.
Structure diagram of RDB.
Figure 3.
u is n 8-dimensional vectors obtained by encapsulating and flattening the data through the PrimaryCaps layer, v1 is a 16-dimensional digital capsule vector after a routing process and v2 is the 32-dimensional vector output at the end of the second routing process. wij is the fully connected weight matrix of the connected vector. The vectors u' and v' are the prediction vectors of u and v1 obtained through a linear hierarchical relationship, the subscripts i, j correspond to the number of vector u and the next layer vector v1, respectively. The parameter bij is the log prior probability from the lower capsule to the upper capsule. The parameter bij is the logarithmic prior probability from the low-level capsule to the high-level capsule and is used to update each cij correspondingly. cij is the coupling coefficient connecting the two layers' vectors before and after, representing the degree of correlation between the i-th vector of the previous layer and the j-th vector of the next layer. Squash is a linear rectification function, which normalizes the results obtained.