I am building an image processing pipeline using OpenCV3 and TBB. When passing the image in a flow graph, it is being altered.
Here is a sample result of the image when passing through a inverse filter:
I wrote a unit test that passes images through a graph and compare the output to the original image:
https://gist.github.com/des0ps/74528673d271e3131c2ea2a2adaf5c8c
This test sometimes passes, and sometimes doesn't.
What is happening here and generally speaking, are cv::Mat supported by TBB?
It seems that the test does not ensure that images being checked are corresponding images,
i.e. the left image has the same sequence number as the right one. This is possible since there
are no restrictions on how fast threads pass the images through the graph. Try to use
'key_matching' buffering policy for join_node.
By the way, TBB is a template library. It means that its structures and algorithms could be
instantiated with any type specified as an argument for a template parameter (of course if that
type satisfies requirements of the structure/algorithm in question). For example, source_node
requires that its output type be DefaultConstructible, that is its instances are created by
calling class default constructor. Such requirements are automatically checked by compiler.
