The problem of runtime

[Hill-Jiang]

@movehand
Hello, can I ask you a question about the running speed of the program?
GOOGLE says that the RAISR runtime is about 10^-2 to 10^-1 seconds.
But I test it and the actual running time of a 512*512 medium sized image on the i7 processor was tens of seconds.
Why the difference between the two running time is so huge?

[movehand]

I think it's simply the method of the actual implementation. This project is simply a rough guide of how RAISR works. So, without unrolling loop, parallelizing, etc. PyCUDA may be a nice option to speed things up!

[BlauerHunger]

It would probably help to get rid of the numpy arrays. Then it would be easy to parallelize the preprocessing on the cpu using multiprocessing.Pool.

[bluewidy]

@BlauerHunger
How can I do that? Can you show me the parallel processing source code file?

[BlauerHunger]

Multiprocessing is explained here: https://docs.python.org/3/library/multiprocessing.html
You can run a function over every element of a list in n processes running in parallel using Pool(n).map(function, list). Since the globals are copies and not shared, sharing data (such as the q and v) must be done either using returns (this might even be possible using the numpy arrays) or this way: https://docs.python.org/3/library/multiprocessing.html#sharing-state-between-processes

Additional efficiency could be gained by adding type information and compiling the project with cython.

[excllent123]

@Shuaibishay Have you compared the result with A+? The result is bad compared with A+.and train takes a week in i7-8700 under div2k dateset.

[hezc13]

@excllent123 Could you offer me your A+ code?I want to compare the results between the two algorithms.You can touch me with [email protected]

[jtoy]

any plans to get this working?