tdsuper / sphericalsuperpixels

superpixels for spherical panorama

Spherical Superpixel

This repository contains the dataset, the evaluation code and the source code of our works on spherical superpixels.

Dataset

Currently, our dataset contains 75 manually annotated spherical panoramas. The original panoramas are selected from SUN360 dataset. Each image is resized to 1024x512 pixels and segmented with an annotation tool based on the application provided by the authors of BSD dataset.

For each spherical panorama, we provide the ground truth segmentations and ground truth boundaries. Some annotation examples are shown below. If you use this dataset in published work, please cite our paper as

@InProceedings{SPSDataset75,
    author   = {Liang Wan and Xiaorui Xu and Qiang Zhao and Wei Feng},
    title    = {Spherical Superpixels: Benchmark and Evaluation},
    booktitle  = {ACCV},
    YEAR   = {2018},
}

Code

Currently, there is only the executable for spherical SLIC algorithm. If you use it in your work, please cite our paper as

@ARTICLE{SLICSTMM, 
    author = {Q. {Zhao} and F. {Dai} and Y. {Ma} and L. {Wan} and J. {Zhang} and Y. {Zhang}}, 
    journal = {IEEE Transactions on Multimedia}, 
    title = {Spherical Superpixel Segmentation}, 
    year = {2018}, 
    volume = {20}, 
    number = {6}, 
    pages = {1406-1417},
}

sphericalsuperpixels's Issues

Hangs at the end of program

Hi, when I run the executable, it prints out "use 1275.36 ms to generate 9 superpixels", and outputs the image and txt file. So it works fine. However, the program doesn't actually end, it hangs (when I run it from the command line, I have to manually terminate it). I'm not sure if you've changed something in the executable that pauses it at the end? This used to work without hanging on my laptop.

I'm using Windows 10.

Let me know if you know of any solutions to this.

Cheers.

Distance types?

What are the possible options for the distance type argument -t? I cannot see it specified in the readme.

From your paper, I see there is Cos-SphSLIC and Avg-SphSLIC. Any others? And how do we specify them? "-t Avg-SphSLIC"? Or does it expect some kind of integer ID? "-t 0"?

More useful output?

Hi, I tested your spherical segmentation, and it works nicely. However, it's an executable, so I cannot modify it or get access to the segmented pixels. The current output is an outline of the segmentation, which is nice for visualisation, but not useful for my actual task (I do not have a reliable way of extracting each segment). A more useful output would be one where each segment has a unique colour ID. That way I can easily use the output as a mask to extract the segments.

Would it be possible to modify the executable to output such an image? Or alternatively provide the source code?