Original RPN2T project paget:
https://github.com/jimmy-ren/RPN2T

We modify the RPN2T tracker by adding flow in sampling strategy and modify the network to siamese network, the paper named Flow Guided Siamese Network for Visual Tracking

This software is implemented using Caffe and part of Faster_rcnn.

This code is tested on 64 bit Linux (Ubuntu 14.04 LTS).

Prerequisites

0. MATLAB (tested with R2014b and R2016a)
1. Caffe (included in this repository external/_caffe/, or you may choose the official version of caffe)
2. For GPU support, a GPU, CUDA toolkit and cuDNN will be needed. We have tested in GTX TitanX(MAXWELL) with CUDA7.5+cuDNNv5 and GTX 1080 with CUDA8.0+cuDNNv5.1.

Compile Caffe according to the installation guideline.

cd $(RPN2T_ROOT)
cd external/_caffe
# Adjust Makefile.config (For example, the path of MATLAB.)
make all -j8
make matcaffe

Compile LK algorithm (using Matlab)

compile

Demo

Run (using Matlab at RPN2T_ROOT Folder)

1. GPU id can changed in the file ./tracking/rpn2t_init_rpn.m;
2. before showing demo, you may need modify dataset directory in ./utils/genconfig.m
3. before showing demo, you should run the command: addpath(genpath('./'));
4. run ./tracking/demo_tracking.m to show demo without GUI. In this way, you only can run with videos form benchmarks;
5. or run FRPN2T_GUI_demo.m to show demo with GUI, in this way, you can run with a video (.avi for example) and choose a target manually.

OTB100