Article Directory
Preface
pysot-toolkit is a tool for evaluating single-target tracking performance. It can calculate Success and Precision under OPE (One Pass Evaluation), as well as VOT (Visual Object Tracking) performance indicators Accuracy, Robustness, EAO, etc. This article is tested under ubuntu system.
1. Environment construction
First enter the pysot-toolkit download command on the command line:
Then install the library package in requirements.txt under pysot-toolkit:
tqdm
numpy
glob
opencv-python
colorama
numba
When installing these packages, you can create a separate virtual environment for pysot-toolkit to use, or you can install missing packages in the existing environment. I installed Colorama directly in the existing environment. After entering the virtual environment, the installation instructions are unified Use pip install xxx . Finally, the terminal command cd to enter the pysot-toolkit/pysot/utils/ file, enter the following command in the path, python3.7 can be changed according to your interpreter version.
After the compilation is successful, you can see the following files under utils. At this point, the environment is set up!
2. File configuration
The environment is set up, the second step is to configure the required files, first give the official website tutorial, only VOT2016 as an example, others are similar.
–Dataset_dir: VOT2016 dataset path
–dataset VOT2016: dataset name
–tracker_result_dir: results path run by your algorithm
–trackers XXX: tracker name XXX
-
--Dataset_dir: This path is only the path of the VOT2016 dataset. Remember to download the VOT2016.json file and store it under the path of the VOT2016 dataset. Click here to download. This file stores the video information of the dataset.
-
--Tracker_result_dir: I only use pysot-toolkit as the evaluation tool, so the results from the algorithm are prepared in advance and placed under this path.
Next, look at the tree structure of this folder
3. Test and comparison
3.1 Single tracker test
The environment is set up and the files are configured. Then you only need to enter the following commands in the terminal in order to evaluate the performance of the algorithm. First, enter the environment you have built, and then enter the evaluation instructions under the pysot-toolkit file in that environment.
The data set path and results path given here use absolute paths, which are more clear and impressive.
3.2 Comparison of multiple trackers
Let's take a look at the folder tree structure first.
The results of all trackers are named after its name. What is a bit confusing here is that there must be a baseline sub-directory to run the evaluation normally. I haven’t read the evaluation source code yet, so it’s not very clear. Comments from friends who want to know After discussion, thank you! The commands and results entered by the terminal during the comparison are as follows: