Testing the 2d tracking to improve performances

[kalwalt]

As discussed in PR #3 I did a profiling test with chrome this is the result:

the most time spent is on _processImage > process function and the wasm child functions.
i leave the profiler file here

CPU-20190915T122749.zip
I would to know what the wasm function are but i don't know how to read them.
I tested the master branch.

[ThorstenBux]

Yeah, we need to compile with debug settings inside emscripten which would leave the names of the C-function inside the wasm file :).

[kalwalt]

You are right i forgot this! i will look into it.

[kalwalt]

i did a profiling with debug symbols with the chrome browser, you can find here:
Profile-20191101T123912.zip

i did it while the image is detected and tracked.

[ThorstenBux]

from the file it looks like the detectAndCompute function takes very long. This is an OpenCV internal function

[ThorstenBux]

I think I should also make one profile run with OpenCV 3 so we can compare the two of them. Would need to setup a reliable OpenCV 3 build pipeline.

[kalwalt]

yes @ThorstenBux i tested with the OpenCV 4, we need to setup the pipeline for OpenCv 3.

[kalwalt]

i think that the first version you let me try was with OpenCV 3, right? So i think shouldn't be too difficult to revert to it. I think, you use it OpenCV 3.4.1, I will check again but i'm quite sure of this, Maybe we could try with other OpenCV 3 versions ? let say 3.5 or 3.6 or other?

[ThorstenBux]

3.4.1 was the one I used, yes. No idea about other 3.x versions, might be worth a shot. But to get quick results we should compare 3.4.1 to 4.x and see if it indeed gets the performance down.

[kalwalt]

ok i will do with the 3.4.1 for now when i have a bit of time.... 😄

[ThorstenBux]

Awesome, thank you :)