Replace matplotlib with much faster OpenCV 3.2 in the show_frame func…

[Greendogo]

…tion.

Fixes Issue #4

I went from ~6fps to ~30fps on my machine.

This still requires changes to the "requirements.txt" file and a python module with OpenCV compiled to use imshow and waitKey must be required (currently the popular but unofficial opencv-python module does not implement them).

Since show_frame no longer uses matplotlib and show_crops and show_scores don't look like they're used at all (didn't check everywhere) it looks like the requirement for matplotlib could be removed along with the references to it in visualization.py and tracker.py (in the imports in that file, and on line 135).

[jvlmdr]

Thanks!

I think it is OK to introduce a dependency on OpenCV. This may also be useful if we want to do e.g. a live demo using a webcam.

[Greendogo]

If I wanted to do a live demo using a webcam (provided I already have a groundtruth for my first frame) with the siamese fc tracker, does anyone have any suggestions regarding where to start?

[Greendogo]

My plan is to attempt to divide Siamfc-tf 's current "tracker" function into three which would be used like so:

tracker = createSiam(params1)
tracker.init(initial_frame, pos_x, pos_y, w, h)
while True:
  frame = camera.read()
  bbox = tracker.update(frame)
  ~<put bbox on frame>~
  cv2.imshow("Siamese Tracker", frame)
  cv2.waitKey(1)

Any advice on parameters or moving things around in the "tracker" function would be most appreciated!

[hayatibis]

I think this line should be

frame_adjusted = np.ndarray(shape=np.shape(frame), dtype=np.dtype(np.uint8))