change model.predict() to model(), notably improve memory leak (issue #697)

[Villyli]

To help improve #697, I use model() replace model.predict() and get notably improvement, which aviod generates a data generator in model.predict().

In my case, for a 1000 times loop, the memory increment is about 50 MiB. it seems that DeepFace.represent() now is capable of 100k loop for most use.

[Villyli]

Besides, u can use a redirect stdout warpper to clean annoying process output

original_stdout = sys.stdout
sys.stdout = open(os.devnull, 'w')

# your codes intended to ignore output in terminal

sys.stdout.close()
sys.stdout = original_stdout

[merasil]

this seems to fix my problem that i had which caused to lock up my app after long runtime when using facenet512 and yunet

[serengil]

thank you for your contribution