This is the code for training of GOTURN tracker implemented inside OpenCV.

Original GOTURN paper:

Learning to Track at 100 FPS with Deep Regression Networks,
David Held, Sebastian Thrun, Silvio Savarese,
European Conference on Computer Vision (ECCV), 2016 (In press)

• Install Caffe and compile using the CMake build instructions: http://caffe.berkeleyvision.org/installation.html

• Install OpenCV

sudo apt-get install libopencv-dev

For GOTURN training, first HDF5 dataset shold be generated. It can be done by simple function call:

buildDB()

By default it generates 10 HDF5 datasets every with 500x10 samples (10 crop samples per image as proposed by authors). As alternative more low-level function can be used:

buildH5Datasets(datasetName, numberOfSamples)

GOTURN training requires a "bvlc_reference_caffenet.caffemodel" for GOTURN network weights inialization:

http://dl.caffe.berkeleyvision.org/bvlc_reference_caffenet.caffemodel

Training is launched by next line:

trainNet()

All hyperparameters are configured in goturnSolver.prototxt, for more details refer Caffe documentation.

In order to visualize results, GOTURN tracker can be tested on test dataset:

buildH5Datasets("D:/ALOV300++/trainDataset.h5", 10);
testNet("goturn_iter_30000.caffemodel");

First command generate a new small test dataset, and second launching a test procedure with visualization.

Also there is pretrained GOTURN model is available in OpenCV_extra repository

https://github.com/opencv/opencv_extra