Lightweight Image Super-Resolution Based on Re-Parameterization and Self-Calibrated Convolution
Algorithm 1
RepSCN function.
Input: the low-resolution image: IL,R, the shallow feature extraction function: fs(.), the number of RepDB: n = 3, the number of SCDB: m=3, the RepDB function: fRepDB( ), the SCDB function: fSCDB( ), the feature fusion part: ff(cat( )), the reconstruction function: frec(.)
Output: the super resolved image: ISR
1:
the shallow feature F0 = fs(ILR)
2:
//Calculate intermediate features Fi and Fj
3:
for i = 1,...,n do
4:
Fi = fRepDB(Fi−1)
5:
end for
6:
for j = n + 1,...,n + m do
7:
Fj = fSCDB(Fj−1)
8:
end for
9:
the aggregated feature Ff = ff (cat(F1,...,Fn,...,Fn+m))