This repository contains several examples of tracker integration for Python, C/C++ and Matlab trackers. Helpers for all languages are provided so that integration is as easy as adding a few lines of code to your tracker source code. For Python example trackers are provided in the python directory, for C/C++ three example trackers are in the native directory and for Matlab an example tracker is in the matlab directory. More information can be found in the documentation.
A NCC tracker example implemented in Python is provided to demonstrate the integration of a tracker with the toolkit. Implementation of the tracker is available in the ncc_tracker.py file. The following examples are available:
-
ncc_singleobject.py: A bounding box tracker run on a single object tracking task.
-
ncc_singleobject_rgbd.py: An example of multi-modal tracking. The main difference compared to
ncc_singleobject.pyis demonstration of how both, the rgb and depth images are obtained during evaluation. This can be used on e.g., VOT-RGBD2020. -
ncc_multiobject_mask.py: A segmentation-based tracker, i.e. segmentation mask is used for initialization and to report the tracking output. This example demonstrates also the multi-object tracking (available since VOTS2023).
-
ncc_singleobject_mask.py: A segmentation-based tracker, i.e. segmentation mask is used for initialization and to report the tracking output. This example demonstrates how a single-target tracker could be run on multi-object dataset, i.e., sequential mode. For example: run a single-target tracker on VOTS2023 multi-target dataset.
-
ncc_multiobject_manager.py: This is a similar demo as
ncc_multiobject_mask.py, just the integration is done slightly more elegant using the VOT manager.
Note: the selection of the demo your integration code should based on, depends on the tracking challenge, dataset specification and your tracker, e.g. single- or multi-object; bounding box or segmentation mask.
Three examples are provided by the toolkit. There are two static trackers (trackers that just report the initial position) written in C and C++ and a NCC tracker written in C++ and using OpenCV library.
All three trackers are using vot.h header that provides integration functions and classes that can be used to speed up the integration process. When compiling the tracker, the wrapper expects that trax.h is available and that the TraX library is found during tracker runtime.
A NCC tracker example, written in Matlab is provided by the toolkit. The tracker uses vot function that generates a communication structure that is then used to communicate with the toolkit. It assumes that traxserver MEX file is available somewhere in the Matlab path.
Note that due to decreasing interest in Matlab as a language of choice in computer vision community, this wrapper currently only supports single object tracking.