Learning to Track at 100 FPS with Deep Regression Networks

1. Github

https://github.com/davheld/GOTURN

Introduction

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.

---

---

2.Installation

1. Install CMake:

   sudo apt-get install cmake

2. Install TinyXML (needed to load Imagenet annotations):

   sudo apt-get install libtinyxml-dev

3. clone GOTURN

   git clone --recursive https://github.com/davheld/GOTURN.git

4. make
   mkdir build
cd build
cmake ..
make

3. 遇到的問題

1. 配置caffe路徑
   需要在 GOTURN/cmake/Modules目錄下修改Caffe_DIR的值：
   ###Set the variable Caffe_DIR as the root of your caffe directory
set(Caffe_DIR /home/dl/caffe)

2. make編譯時遇到以下編譯錯誤：
   fatal error： caffe/proto/caffe.pb.h:No such file or directory #include 'caffe/proto/caffe.pb.h'
   解決：
   切換到caffe根目錄：
   （1） proto src/caffe/proto/caffe.proto –cpp_out=.
   【–cpp_out 指定生成c++的代碼並放置在當前路徑中 】
   (2) mkdir include/caffe/proto
   (3) mv src/caffe/proto/caffe.pb.h include/caffe/proto

4. Pretrained model

bash scripts/download_trained_model.sh

5. Tracking Performance

bash scripts/show_tracker_test.sh /home/dl/Downloads/vot2015


---

---

---

Reference

1. https://github.com/davheld/GOTURN

2. http://blog.sina.com.cn/s/blog_721a75e50102wfig.html