Downloads:
From Add-On in Matlab setup Parallel Computing Toolbox
MatConvent: https://www.mathworks.com/matlabcentral/fileexchange/47811-vlfeat-matconvnet
If you expireince error in compiling matconvnet use this vl_compilenn('EnableGpu', true, 'cudaMethod', 'nvcc','verbose', '1')
VS2015 https://my.visualstudio.com/Downloads?q=Visual%20Studio%202015%20with%20Update%203
Install GPU
http://www.linuxandubuntu.com/home/how-to-install-latest-nvidia-drivers-in-linux
or https://askubuntu.com/questions/149206/how-to-install-nvidia-run
I installed cuda by
https://www.youtube.com/watch?v=FK1y7XQuhp0
Latest vesrsion of Nvidia driver https://help.ubuntu.com/community/BinaryDriverHowto/Nvidia
CUDA & CUDNN
- https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html#installcuda-windows
- https://developer.nvidia.com/rdp/cudnn-download
Choose the suitable version for your Windows, GPU.
-
Visual Studio 2017 doesnt work with Matlab 2017b in case of installing Matconvent
-
Instead use VS2015
-
Dont forget to add CUDNN to CUDA9.0
-
Becarful after you add CUDNN python doesnt work.
-
To explore which compiler you want to use (VS2017, VS2015) in case you have both just type in matab mex -setup it allows you to choose one.
- I downloaded TRAX from here https://github.com/alessiodore/trax
- Replace every xrange by range if any
- Vot toolkit/ tracker/ System_wrapper.m I added this line before line#176
#Instead of# :
[status, output] = system(tracker.commandcomand, '');
#ADD#
command= string(sprintf('python %s', tracker.command));
[status, output] = system(command, '');
-
In Trax.py add import fuctools
-
At line 392
#instead of#assert(reduce(lambda x,y: x and y, [isinstance(p,tuple) for p in points], False))
#ADD#
assert(functools.reduce(lambda x,y: x and y, [isinstance(p,tuple) for p in points], False))
-
If map function doesnt have length add list before map.
-
The same for all trax files (e.g. Images, Region,...)
-
For CCOT Tracker use this toolkit version https://github.com/votchallenge/vot-toolkit/tree/64da0655cf973ef4a32923c72d5c423f908325a5
-
How to install CCOT tracker https://github.com/martin-danelljan/Continuous-ConvOp
-
ECO trackerhttps://github.com/martin-danelljan/ECO When you run it and gives you error message Make sure you integrate CUDNN (Just add folders to CUDA) and also be sure that IMAGENET it had been downloaded
-
When you download any tracker from VOT-Challenge http://www.votchallenge.net/challenges.html Don't forget to check the MATCONVENT version (depends on your OS and GPU) they use otherwise delete their MATCONVENT and put yours
Other issues
After installing CUDNN when you run Matlab (Run_expierment.m in Vot-workspace) it will give you erro
to fix it you have to remove CUDnn files form (bin, lib, include folders).
Don't run Matlab in VOT while python Anaconda-Spyder is running).
Number of folders (seuqences in VOT/results folder for each tracker) should be the same for other trackers