Neo-PSP-CRF

This repository is a new implementation of the system elaborated in this paper: Advances in Semantic Image Segmentation using Probabilistic Superpixel-based Dense Conditional Random Field, shortened as Neo-PSP-CRF. To grasp the core concept quickly, please read the paper's abstract.

How to execute this code

(1) Image processing is a key requirement of this project, so openCV needs to be installed first. Overall project management is done using Cmake.

(2) Download the code, extract it and navigate to the root directory of the project. The main.cpp is the core file and is dependent on some utility functions. Hence, compilation of some static library files is required.

(3) Navigate to the cnpy-master directory in src, where you need to obtain libcnpy.a using CMakeList.txt. It is used for reading Probability files of CNN into the memory to improve read speeds. The libcnpy.a file can be found in the build directory.

(4) In the densecrf_lib directory, obtain libdensecrf.a using CMakeList.txt, this function is used for the CRF API.

(5) Likewise, in the lbplib directory, obtain libopencv_lbp.a using CMakeList.txt for the LBP API.

(6) Finally, from the SLICO directory in src, obtain libSLIC.a by the CMakeList.txt, This function is used for the SLIC API.

After completing steps 1-6, you will have 4 .a files. Copy these to the lib directory. The downloaded code will already have a lib directory, but you need to recompile for your machine.

(7) Now, return to the root directory, open CMakeLists.txt and modify paths according to your environment. Then compile the executable file.

If it executed properly, for a single image, the arguments should look like this: ./neo-psp-crf a.jpg a.npy out.png 0 3 30 2 3 2 1 12.

(8) In the root directory, you'll find the eval.py file which is a Python file used for the automatic execution of many images. Modify the file paths before running.

Citation

Please cite Neo-PSP-CRF in your relevant publications:

@article{Neo-PSP-CRF,
     Author = {Zhang, Liang and Li, Huan and Shen, Peiyi and Zhu, Guangming and Song, Song and Shah, Syed Afaq Ali and Bennamoun, Mohammed and Zhang, Li},
     Journal = {IEEE Access},
     Title = {Improving Semantic Image Segmentation With a Probabilistic Superpixel-Based Dense Conditional Random Field},
     Year = {2018}
     volume={6},
     Number={1},
     Pages={15297-15310},
  }

Participate in discussions via Email: [email protected], Li Huan