Fast Computation of Content-Sensitive Superpixels and Supervoxels Using Q-Distances

This repository contains the source code for the ICCV 2019 paper Fast Computation of Content-Sensitive Superpixels and Supervoxels Using Q-Distances by Yong-Jin Liu.

Running Environment

The developing environment we used is Microsoft Visual C++ 2012, and we have tested our executable files on 64-bit MS Win7, Win 8 and Win 10. We recommend users to compile the process in x64 and Release version.

subprogram

We define many functions for different tasks in MainProcess.h, and the corresponding input parameters are defined in main.cpp. Users can comment and uncomment the code to compile different versions for different tasks.

Superpixels for RGB images

You can see the function superPixel_rgb_iter_mt, The command is as follows (rename the executable file as qd-CSS_rgb.exe):

qd-CSS_rgb.exe  input_folder  output_folder  number_of_superpixels  iter_max

input_folder: the filepath of input folder containing RGB images which would be segmented.The types can be JPG, PNG or BMP. output_folder: the folder filepath contains corresponding superpixels result BMP images. number_of_superpixels: the number of superpixels you want to generate, which should be integer. 200-700 recommended. iter_max: the max number of interation, which should be integer. 5-20 recommended.

Example:

qd-CSS_rgb.exe examples_input examples_output 300 10

Superpixels for RGBD images

You can see the function superPixel_rgbd_iter_mt, The command is as follows (rename the executable file as qd-CSS_rgbd.exe):

qd-CSS_rgbd.exe  rgb_image_folder  depth_image_folder  output_folder  number_of_superpixels  iter_max

rgb_image_fold: the filepath of input folder containing RGB images which would be segmented.The types should be PNG format. depth_image_folder: the folder filepath contains corresponding depth input which should also be PNG format. output_folder: the folder filepath contains corresponding superpixels result BMP images. number_of_superpixels: the number of superpixels you want to generate, which should be integer. 200-700 recommended. iter_max: the max number of interation, which should be integer. 5-20 recommended.

Example:

qd-CSS_rgbd.exe images depth output 300 10

Supervoxels for RGB videos

You can see the function superVoxel_submit, The command is as follows (rename the executable file as qd-CSS_sv.exe):

qd-CSS_sv.exe  input_folder  output_folder  number_of_supervoxels  iter_max

input_folder: the filepath of input folder containing video frames which would be segmented.The types can be JPG, PNG or BMP. output_folder: the folder filepath contains corresponding supervoxels result png images. number_of_supervoxels: the number of supervoxels you want to generate, which should be integer. 200-5000 recommended. iter_max: the max number of interation, which should be integer. 5-20 recommended.

Example:

qd-CSS_sv.exe sv_input sv_output 500 10

Additional notes

Citation:

@InProceedings{Ye_2019_ICCV,
author = {Ye, Zipeng and Yi, Ran and Yu, Minjing and Liu, Yong-Jin and He, Ying},
title = {Fast Computation of Content-Sensitive Superpixels and Supervoxels Using Q-Distances},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month = {October},
year = {2019}
}