Face cluster

[hauck-jvsh]

Group similar faces and create a bookmark for each group. Maybe this could be enhanced if there is an API to change the itens, saving the group of each face in a metadata.
Closes #1475

[lfcnassif]

Thank you @hauck-jvsh! Yes, #1185 is planned, I think it is needed for this and other features.

[hauck-jvsh]

I found the reason it is taking a lot of time during the process and not so much otherwise.
if pingExternalProcess(proc): #from threading import Thread #t = Thread(target=log_stderr, args=(proc,)) #t.daemon = True #t.start() return proc
Comment those lines and the times will be:

Time(s) to extract 36895 faces 6.684337139129639
Time(s) to group similar faces: 9.375169515609741

I think that it is creating a huge amount of threads and they are overwhelming the JVM.

[hauck-jvsh]

Actually it speed up the hole processing in this case. I think that the problem occurs when there is a lot of erros during the face extraction.

[lfcnassif]

Hum... very interesting! I think that block was created to read stderr of the created external processes to avoid blocking them if they write too many errors to stderr, possibly there is a better way to do this, maybe reusing the same Thread to read stderr of several processes, not sure how to implement it in python...

[lfcnassif]

Possibly the huge slowdown was caused by python GIL...

@fmpfeifer, do you have any suggestion on how to ignore stderr of external processes started from python without using threads and without blocking the main thread that started the external processes?

[lfcnassif]

Actually that block is used to log stderr, not ignore it.