Performance issue on post-processing

[tucan9389]

Now:
Inference -> convert to primitive array from mlarray -> draw the converted array

Suggest:
Inference -> draw the mlarray