Use of pcolormesh instead of imshow

[jurjendejong]

Switching to new visualisation method for higher framerate.

A test has been done with 1000 frames of a changing 100 x 100 matrix. Different engines have been tested. Script is checked in in the Bastei branch in the 'playground' folder as 'compare_plotting_functions.py'.

Module	Methode	Frequentie (s-1)
matplotlib	imshow	8.16
matplotlib	imshow with set_data	10.27
matplotlib	pcolormesh (with set_array)	21.96
matplotlib	pcolormesh for larger window	12.63
vispy	App, canvas, image	61.44
cv2	imshow	64.02

3-4-2018: Update using cv2 and changing the timing from time.process_time() to time.time()

[jurjendejong]

• Update on visualisation: pcolormesh did not prove to be much of a gain (somehow).
• Rewritten the code to have one imshow for the combined data of waterdepth+kinectheight+dotjes, switching from a data arrays (3 arrays of m x n cells) to one color array (m x n x 4)
• Add test with cv2 as another plotting library (also imshow)
• Switch to cv2?

[jurjendejong]

Finished the upgrade to cv2, however, the visualisation has gotten a lot slower. I suspect that the actions of array transformation are getting too expensive to achieve the high framerate. Profiling is required.

[jurjendejong]

timestep	1	2	3	4	5	6
update_kinect_height	0.010	0.097	0.004	0.003	0.007	0.004
update_height_cells	0.004	0.004	0.003	0.003	0.003	0.003
update_waterdepth	0.018	0.015	0.008	0.012	0.010	0.010
update_height_cells	0.004	0.006	0.003	0.004	0.003	0.002
update_waterdepth	0.022	0.014	0.012	0.015	0.009	0.008
update_uv	0.009	0.024	0.006	0.009	0.006	0.006
seed_lic	0.005	0.004	0.004	0.004	0.004	0.004
update_lic	0.085	0.071	0.053	0.069	0.048	0.047
vis_kinect_height	0.019	0.018	0.013	0.014	0.018	0.017
combine_images	0.038	0.043	0.035	0.046	0.029	0.029
vis_waterdepth	0.025	0.024	0.017	0.027	0.017	0.019
combine_images	0.032	0.049	0.032	0.029	0.034	0.039
vis_lic	0.000	0.000	0.000	0.000	0.000	0.000
combine_images	0.038	0.040	0.028	0.027	0.037	0.025
sum of above	0.309031	0.409041	0.21802	0.262026	0.225023	0.213022
update	0.335033	0.40104	0.222022	0.269027	0.231023	0.221022

[jurjendejong]

After further investigation of the script and the table above, i've come to the conclusion that matrix/array computations in python take most of the time. Not the model, not the plotting, not the video analysis, but just some basic matrix multiplications.

[jurjendejong]

Suggestions by Fedor:

1. Use numpy MKL version
2. Merging of images using CV2
3. Making canvas with matplotlib, but plotting with cv2