Python Testing Callback : Way to get test output

[VasLem]

Hey guys. I have dived in very deep waters of Caffe Python.  I have to perform a multilabel classification. Apart from that, I am training a network using multiple gpus.
I want to get the mAP during training.
For this reason I have added test_iter and test_interval in the solver prototxt, in order to be able to process network output using some Python Code after each interval and to get finally the wanted precision, by using the callbacks provided. Unfortunately I am doing a dissimilarity check to get the mAP, which is not implemented as a Caffe layer. The problem is that I do not have a way to get the testing output out of the network after this interval, so I am forced to test the network AGAIN, by feeding each input from the testing set and using the net.forward() command, so that to save the output to a list and then perform the dissimilarity check, thus delaying by 5 mins per epoch, for let's say 50 epochs, thus getting a high ending total delay. Can someone suggest a way not to have to retest the network? I have thought of a Python layer, but I do not know even then the way to get it to return the mAP for all the input test data.