Antrack evaluation on Wu. et al (2013), Wu. et al (2015) tracking benchmark

See original README. Provides functions to run tracker evaluation in parallel.

Running original

Usage

• Default (for all trackers, all sequences, all evaltypes(OPE, SRE, TRE))
  • command : python run_trackers.py
  • same with
• For specific trackers, sequences, evaltypes
  • command : python run_trackers.py -t "tracker" -s "sequence" -e "evaltype"
  • e.g : python run_trackers.py -t IVT,TLD -s Couple,Crossing -e OPE,SRE)

Libraries

• Matlab Engine for python (only needed for executing matlab script files of trackers)

  http://kr.mathworks.com/help/matlab/matlab-engine-for-python.html

• matplotlib

• numpy

• Python Imaging Library (PIL)

  http://www.pythonware.com/products/pil/

Running parallel

Original benchmark evaluation has 2 drawbacks. First, it only evaluates the trackers sequentially. Second, results are saved only in the end. run_trackers_cached_parallel.py allows to save evaluation for each video independently; it also allows to run the evaluation in parallel. Syntax is as follows:

python run_trackers_cached_parallel.py -t RobStruck -e OPE -p 16 # runs evaluation of the RobStruck tracker using OPE protocol with 16 threads

# it is possible to run more than one tracker

python run_trackers_cached_parallel.py -t RobStruck,ObStruck,MBestStruck -e OPE -p 16

Limitations

• Sequences have to be downloaded either manually or using original benchmark (original). Once its done a symlink ./data/ -> place where sequences are downloaded would suffice.

• Evaluations on SRE,TRE have different parameters than the MATLAB toolbox. This results that during evaluation results needed to be trimmed.