/CRTrees

Hierarchical Superpixel Segmentation by Parallel CRTrees Labeling. IEEE TIP 2022

Primary LanguageCudaGNU General Public License v3.0GPL-3.0

Hierarchical Superpixel Segmentation by Parallel CRTrees Labeling

Please cite the [paper] if you find it useful

@ARTICLE{9819438,
author={Yan, Tingman and Huang, Xiaolin and Zhao, Qunfei},
journal={IEEE Transactions on Image Processing},
title={Hierarchical Superpixel Segmentation by Parallel CRTrees Labeling},
year={2022},
volume={31},
number={},
pages={4719-4732},
doi={10.1109/TIP.2022.3187563}}

Workflow

CRTREES

Dependency

  • CUDA >= 6.0
  • OpenCV >= 3.0

Usage

Compile the code

unzip lib_eval.zip // this is for evaluation on datasets
bash build.sh

Test on a single image (the first run takes time to load the GPU)

bash run_img.sh

Video streams (a web camera is required)

bash run_video.sh

CRTREES can achieve 200+fps for 480P video streams on a Titan Xp GPU. Faster speed can be achieved if the GPU version of OpenCV is used.

Benchmark

The superpixel benchmark (https://github.com/davidstutz/superpixel-benchmark) shall be put in the same level directory as this project.

bash bench_superpixels.sh BSDS500

The same results as in the paper can be obtained.