Opencv_contrib: Change psr_threshold parameter for CSRT-tracker in Python

[AlexeyAB]

A suggestion to add the ability to change the psr_threshold parameter in Python-wrapper for CSRT-tracker, which is currently not available: https://stackoverflow.com/questions/54785230/csrt-algorithm-not-updating-target#comment96385997_54790163

System information (version)

• OpenCV => 4.1.0
• Operating System / Platform => Any
• Compiler => Python 3.x

Detailed description

Steps to reproduce

tracker = cv2.TrackerCSRT_create()

[8talha]

Any solution for this? (A separate question) Moreover do this CSRT has cuda 8.0 implementation? I am working on Jetson Tx1 with this algorithm and currently I have installed opencv with cuda enabled but fps is still low it gives fps of 15-20 and GPU performance of max 50%.

[mshabunin]

Try to replace CV_EXPORTS in Params class with CV_EXPORTS_W_MAP. If it does not help, try to move this class outside of TrackerCSRT.

[unrelatedlabs]

You can update the parameters with .write() on the tracker.
use .read() to read out the defaults

[AlexeyAB]

@unrelatedlabs Thanks! Can you provide an example?

[unrelatedlabs]

Sure, it was a little tricky to figure it out initially.

Write

cstr = cv2.TrackerCSRT_create()
fs = cv2.FileStorage("settings.yaml",cv2.FILE_STORAGE_WRITE) 
cstr.write(fs)
fs.release()

Read

fs_settings = cv2.FileStorage("settings.yaml",cv2.FILE_STORAGE_READ)
tracker = cv2.TrackerCSRT_create()
tracker.read(fs_settings.root())

[GreenVegetables123]

@unrelatedlabs，I add psr_threshold in the setting.yaml and chage the value of it,but it do not work

[unrelatedlabs]

@unrelatedlabs，I add psr_threshold in the setting.yaml and chage the value of it,but it do not work

You'll have to experiment with the values. the psr_threshold value does not have any real world sense. it's an arbitrary value that highly depends on the image and bbox.