最近在拜读RBG大神的faster-rcnn源码时发现他用了多进程去分阶段处理神经网络，原因如下：
# --------------------------------------------------------------------------
# Pycaffe doesn't reliably free GPU memory when instantiated nets are
# discarded (e.g. "del net" in Python code). To work around this issue, each
# training stage is executed in a separate process using
# multiprocessing.Process.
# --------------------------------------------------------------------------
大致意思是pycaffe在安装网络后当不再需要使用该部分网络时，不能靠谱的释放GPU显存资源。为解决这个问题，每一个训练阶段都做一个独立的进程去执行，也就用到了多进程。
python的多进程示例如下：
import multiprocessing as mp def function_name():
do sth.
queue.put({'xxx': yyy}) mp_queue = mp.Queue() p = mp.Process(target=function_name, kwargs=dict_ur_mp_kwargs)
p.start()
output = mp_queue.get()
p.join() #其他进程
p = mp.Process(target=function_name, kwargs=dict_ur_mp_kwargs)
p.start()
output = mp_queue.get()
p.join()
#.......