Synopsis

KSPTrack is a method for the segmentation of video and volumetric sequences with sparse point supervision.

Installation

This software depends on the following independent components that you will have to install first. Both make use of the C++ boost library. The installation procedure are given in the respective github repositories as well as here for convenience.

We also provide a docker image that includes all requirements below at lejeunel/boost.

SLIC Supervoxels

simple and efficient supervoxels.

git clone https://github.com/lejeunel/SLICsupervoxels
cd SLICsupervoxels
mkdir build
cd build
cmake ..
make
python3 src/setup.py install

Edge-disjoint K-shortest paths

C++ implementation. Uses the boost graph library.

git clone https://github.com/lejeunel/boost_ksp
cd boost_ksp
mkdir build
cd build
cmake ..
make
python3 src/setup.py install

Install the whole thing

Once both external dependencies are installed, procede to the current package:

git clone https://github.com/lejeunel/KSPTrack
cd KSPTrack
pip install .
pip install -r requirements.txt

Usage

All parameters used in this program are set in cfgs/cfg.py.

We provide two files depending on the availability of GPU:

• single_ksp.py: Uses a pre-trained VGG16 for feature-extraction. It requires no GPU.
• single_ksp_gpu.py: Trains and extracts features from a U-Net. It requires a GPU.