Visual Tracker Benchmark

I'm using benchmark V1.0 to show the results of tracking, and after downloading the code, there's already a variety of tracker results on different datasets.

If you want to verify the effect of your algorithm on this data set, it is actually very simple:

The direct approach:

First copy the code to otb_root_path/tackers/this folder, you will find that there are several algorithms in the Code folder.

The second step is: to OTB_ROOT_PATH/UTIL/CONFIGTRACKERS.M, add the name of your code to the TRACKERS1 structure, for example: struct (' name ', ' KCFDP ', ' namepaper ') , ' KCFDP '),...。

The third step is: Run the MAIN_RUNNING.M function.

For specific examples, refer to: Https://github.com/masa-nudt/KCFDP

This author describes how to run the author's own algorithm with OTB

Indirect method:

The stupid way to do this is: when I need to verify my results, I just need to save the results of my own algorithm as the mat data that the author puts in the results folder.

The following is a detailed talk about:

Opening a folder will reveal many subfolders and four main functions, followed by DRAWRESULTBB.M, GENPERFMAT.M, MAIN_RUNNING.M, and PERFPLOT.M functions.

For the Mainrunning function is the main function, this function is able to divide the video into 20 segments to verify the Tre. At the same time, the Ground_truth of the first frame can be transformed (altogether 12 kinds) to verify the SRE. Finally, you can add your own code in this, so that you can save the tracking results in a specific form, in short, to achieve all the effects.

But want to draw various tracker algorithm in the image effect (bounding boxes), can call drawresultbb.m function, can realize: drawing bounding boxes (BBs) of different trackers on Each frame
If you want to draw a variety of precision curves, you can call the Perfplot function, this function calls the Genpermat function to get the required precision data and error rate (DP,OP), and then put in the Perfmat\overall\ file, the picture saved in Figs\ The Overall\ folder.
At the same time, the author draws the results of 10 methods in 29 tracker methods, using ranknum=10 as the parameter to set, from which the first 10 results are the best. Rankingtype adopts the method of AUC.
Another important parameter to note is Evaltypeset = {' OPE '};%{' SRE ', ' TRE ', ' OPE '};
As with DRAWRESULTBB, Genperfmat is accessing the results/folder when generating the data, so we need to save the results in a standard format and put them in the corresponding folder.

Then say the function of each folder: The Anno folder is a ground_truth of 51 test videos. Figs is used to store the picture of the above-mentioned DRAWRESULTBB.M picture. Initomit is a frame in which each video cannot be initialized, and these frames are generally target-blurred. Perfmat is used to store genpermat results. Results stores the results of different algorithms on different datasets that the authors have run well. The Rsteval store is a number of evaluation functions. TMP stores temporary variables, trackers store this various tracking algorithm source code or EXE file. The functions in Util are the most useful and most likely to be modified. such as CONFIGSEQS.M and CONFIGTRACKERS.M functions.

For the Util folder we need to measure those data sets to go to the function configseqs to take the modified parameters, if you want to compare those algorithms to the Config

Visual Tracker Benchmark