My old libsrcnn experimentally updated for doing scaling with stepping sequence in multiply ratio 2.0 for solve some issues belong to original source.
ShuWang’s OpenCV codes was just seems to belong to matlab, and I had advanced code with OpenMP then native C++ that runs all almost platform with g++ (and optional OpenMP for accelerating processing speed).
But original source codes contained trained data for maximum 2x multiply, So I had advanced some looping mechanism to multi steps by dividing maximum size by double.
Heres’ some tests with down sized famous butterfly in 64px.
Source image : butterfly_64px.png
Each scaling results comaring before and latest with step-scaling.
muliply 1.9x, SRCNN and SRCNN step-scaling (new)
multiply 2.0x, SRCNN and SRCNN step-scaling (new)
multiply 3.0x, SRCNN and SRCNN step-scaling (new)
multiply 5.0x, SRCNN and SRCNN step-scaling (new)
multiplay 7.0x, SRCNN and SRCNN step-scaling (new)
Sources:
Step-scaling currently availed in my github repo.
Supports Windows(MinGW-W64, MSYS2), Linux, MacOSX including armhf, aarch64 debian embedded linuxes.