VOT-2016 代码评测工具的使用说明

VOT-2016 代码评测工具的使用说明

2018-10-14 09:37:04

VOT-2016 官网：http://www.votchallenge.net/vot2016/

评测代码链接：https://github.com/votchallenge/vot-toolkit

用到的工具: https://codeload.github.com/votchallenge/trax/zip/master

参考博文：

https://blog.csdn.net/sgfmby1994/article/details/78776465

https://blog.csdn.net/yao1131/article/details/78783236

Some Important Files: 

1. Pre-trained pyMDNet models (pre-trained on VOT or OTB dataset and testing for OTB and VOT, respectively):

　　链接: https://pan.baidu.com/s/1ImnL8HHLdzgt6JyMYhk0YQ 提取码: bafq

2. pyMDNet code running successfully on VOT-2016 dataset:

　　链接: https://pan.baidu.com/s/11yTqfUSq8FmzdRviWITr7Q 提取码: c54h

　　warning:you need to install pyTorch 2.0 according to:

sudo pip install http://download.pytorch.org/whl/cu80/torch-0.2.0.post3-cp27-cp27mu-manylinux1_x86_64.whl

1. 第一步执行：toolkit_path.m

2. 第二部执行：workspace_create.m

>> workspace_create
Select one of the available experiment stacks:
1 - test
2 - vot2013
3 - vot2014
4 - vot2015
5 - vot2016
6 - vot2017
7 - vot2018
8 - votlt2018
9 - vottir2015
10 - vottir2016
Selection: 5
Input an unique identifier for your tracker: attentionMDNet
Is your tracker written in any of the following languages?
1 - "Matlab"
2 - "Python"
3 - "C/C++"
4 - "Octave"
5 - "None of the above"
Selected option: 1
Verifying native components ...
使用 'Microsoft Visual C++ 2013 Professional' 编译。
MEX 已成功完成。
使用 'Microsoft Visual C++ 2013 Professional' 编译。
MEX 已成功完成。
使用 'Microsoft Visual C++ 2013 Professional' 编译。
MEX 已成功完成。
使用 'Microsoft Visual C++ 2013 Professional' 编译。
MEX 已成功完成。
使用 'Microsoft Visual C++ 2013 Professional' 编译。
MEX 已成功完成。
使用 'Microsoft Visual C++ 2013 Professional' 编译。
MEX 已成功完成。
使用 'Microsoft Visual C++ 2013 Professional' 编译。
MEX 已成功完成。
使用 'Microsoft Visual C++ 2013 Professional' 编译。
MEX 已成功完成。
使用 'Microsoft Visual C++ 2013 Professional' 编译。
MEX 已成功完成。

***************************************************************************

The VOT workspace has been configured
Please edit the tracker_attentionMDNet.m file to configure your tracker.
Then run run_test.m script to make sure that the tracker is working.
To run the experiments execute the run_experiments.m script.

***************************************************************************

【注意】在这个过程中，有可能遇到 trax 的错误:

Downloading TraX source from: https://github.com/votchallenge/trax/archive/master.zip

Please wait ... Unable to unpack TraX source code.

这个时候，我们可以手工下载解压这个软件，放到下图这个路径下：

要么就是你当前的 gcc 版本不匹配导致的，出现如下所示的错误，建议你将 gcc 换成 gcc-4.8 版本，然后重新编译就可以了。怎么换？建议看看这个博文：https://www.cnblogs.com/wangxiaocvpr/p/5385961.html

>> run_experiments
Initializing workspace ...
Checking for toolkit updates on GitHub.
Verifying native components ...
Loading sequences ...
Testing TraX protocol support for tracker run_vot.
Tracker execution interrupted: Invalid MEX-file '/home/vot-toolkit/native/traxclient.mexa64': /usr/local/MATLAB/R2017a/bin/glnxa64/../../sys/os/glnxa64/libstdc++.so.6: version `GLIBCXX_3.4.21' not found (required by /home/vot-toolkit/native/traxclient.mexa64).
TraX support not detected.
Error using tracker_load (line 128)
Tracker has not passed the TraX support test.

Error in run_experiments (line 8)
tracker = tracker_load('run_vot');

这个步骤执行完毕后，会生成几个文件，如下所示：

3. 打开 tracker_attentionMDNet.m，我们可以看到默认的代码如下：

change the code into the following style:

step-1. command the second line;

step-2. change the line 17 according to your own tracker and path. and do not forget to add a '/' at the end of the file path.

4. open the run_experiments.m  and run this script. It will begin to download the VOT-2016 dataset, just hold on ...

>> run_experiments
Initializing workspace ...
Checking for toolkit updates on GitHub.
Verifying native components ...
Loading sequences ...
Downloading sequence dataset "VOT2016 Challenge" with 60 sequences.
Downloading sequence "bag" ...
Downloading sequence "ball1" ...
...

when the dataset is downloaded, it shown me the following errors:

Testing TraX protocol support for tracker attentionMDNet.
Tracker execution interrupted: Unable to establish connection.
TraX support not detected.
Error using tracker_load (line 127)
Tracker has not passed the TraX support test.

Error in run_experiments (line 8)
tracker = tracker_load('attentionMDNet');

according to this blog:  https://blog.csdn.net/HUSTbest_/article/details/80248610

==>>> Ok, now let's compile this tracker first and check if it can success? yes, it indeed successfully run.

跑别人的算法成功了，但是自己的算法，还是不行啊。为什么呢？我们再仔细看下这个成功运行的 tracker，这个文件夹里面有一个 vot.m 函数，但是我们自己的跟踪算法文件夹中，却没有这个！god, damn it!

OK，知道是怎么回事了，就可以拷贝一份到我们自己的算法文件夹中就行了。执行后，发现可以正常进行 Trax 的通信了。

但是，这只是其中的一个步骤，我们必须得修改我们的代码，以使得满足 VOT 评测的格式以及运行方式。来，深呼吸，一步一步来：

reference blog: https://blog.csdn.net/aiqiu_gogogo/article/details/79454997

But we still find following errors when rewrite the code according to the reference tracker: 

>> run_experiments
Initializing workspace ...
Verifying native components ...
WARNING: No configuration for tracker VOT_mdnet found
93 context = iterate(experiments, trackers, sequences, 'iterator', iterator, 'context', context);
Experiment baseline
Tracker VOT_mdnet
Sequence bag
Repetition 1
Tracker execution interrupted: First argument must be a string
Error using traxclient
First argument must be a string

Error in tracker_run (line 78)
data = traxclient(tracker.command, callback, ...

Error in experiment_supervised (line 71)
data = tracker_run(tracker, @callback, data);

Error in tracker_evaluate (line 57)
[files, metadata] = experiment_function(tracker, sequence, directory, parameters, scan);

Error in workspace_evaluate>execute_iterator (line 128)
tracker_evaluate(event.tracker, event.sequence, event.experiment);

Error in iterate (line 69)
context = iterator(event, context);

Error in workspace_evaluate (line 93)
context = iterate(experiments, trackers, sequences, 'iterator', iterator, 'context', context);

Error in run_experiments (line 10)
workspace_evaluate(tracker, sequences, experiments);

==>>  How to solve ???  This maybe caused by your own code. you need to re-write the code and make it compactable with VOT evaluation tools.

When evaluation, you just run the run_analysis.m:

% This script can be used to perform a comparative analyis of the experiments in the same manner as for the VOT challenge
% You can copy and modify it to create a different analyis
addpath('/media/wangxiao/49cd8079-e619-4e4b-89b1-15c86afb5102/aaai_2019/aaai2019_attentionTracking/vot-toolkit'); toolkit_path; % Make sure that VOT toolkit is in the path
[sequences, experiments] = workspace_load();
% error('Analysis not configured! Please edit run_analysis.m file.'); % Remove this line after proper configuration
trackers = tracker_list('run_vot_V2'); % TODO: add more trackers here
workspace_analyze(trackers, sequences, experiments, 'report_AAAI2019_pyMDNet_AttentionV2', 'Title', 'Report for vot2016');

Then you can find the generated evaluation report as shown in above figures.