GOTURN_matconvnet

A matlab implementation of GOTURN. Rather Fast in CPU(20FPS).

GOTURN appeared in this paper:

Learning to Track at 100 FPS with Deep Regression Networks, European Conference on Computer Vision (ECCV), 2016

GOTURN addresses the problem of single target tracking: given a bounding box label of an object in the first frame of the video, we track that object through the rest of the video.

Here is a brief overview of how GOTURN works:

Pretrained model

Download from google drive or BaiduYun

or

Convert caffe models to matconvnet by yourself.

• Download caffemodel

cd <GOTURN_matconvnet>
cd model
wget http://cs.stanford.edu/people/davheld/public/GOTURN/trained_model/tracker.caffemodel

• MATLAB

cd model
Converter_caffe_matconvnet
make_goturn_net

Track

• Compile matconvnet

cd <GOTURN_matconvnet>
git clone https://github.com/vlfeat/matconvnet.git
cd matconvnet
run matlab/vl_compilenn ;

• Download VOT2015

cd <GOTURN_matconvnet>
cd data
wget http://data.votchallenge.net/vot2015/vot2015.zip
unzip -n vot2015.zip -d ./VOT15

• Track in MATLAB

cd <GOTURN_matconvnet>
cd track
show_tracker_test();

VOT2015 speed

**Ubuntu16.04 MATLAB2016b matconvnet-1.0-beta24 **

CPU: Intel(R) Xeon(R) CPU E5-2630 v4 @ 2.20GHz

GPU: GTX 1080ti

And more?

If you have problem, email @foolwood.