/superpixels-generator

The superpixel generator for the support image format

Primary LanguagePythonMIT LicenseMIT

Superpixels Generator

The superpixel generator for the support image format.

The fast way to generate superpixel with target images. It defaults run multiprocessing for the single node. Also, it supports distributed multiprocessing for multi nodes. If using distributed multiprocessing for multi nodes, the script will automatically avoid generating duplicate superpixels. Therefore, it is safe to open one node run the following code, and open another node to run the same argument if using the same config. By the way, the superpixel will save as the .png format.

Introduction

As explained by David Stutz, "Superpixel group perceptually similar pixels to create visually meaningful entities while heavily reducing the number of primitives for subsequent processing steps. As of these properties, superpixel algorithms have received much attention since their naming in 2003. By today, publicly available superpixel algorithms have turned into standard tools in low-level vision." David Stutz et al., "Superpixels: An Evaluation of the State-of-the-Art", CVIU (2017)

Requirements

pip install -r requirements.txt

Configuration

By using factory design pattern, you can configure your method in file config_superpixel.json.

Usage

usage: superpixel.py [-h] -i INPUT -o OUTPUT -m METHOD [--config]
                     [--multiprocessing-distributed]

Superpixels generator

optional arguments:
  -h, --help            show this help message and exit
  -i INPUT, --input INPUT
                        please input the path of image folder e.g.
                        "image_folder/train"
  -o OUTPUT, --output OUTPUT
                        please output the path of image folder e.g.
                        "image_folder/train_sp"
  -m METHOD, --method METHOD
                        please select superpixel method: "watershed", "quick",
                        "fz", "slic"
  --config              use config in file ./config_superpixel.json to
                        generate superpixel based on select method
  --multiprocessing-distributed
                        use multi-processing distributed to generate

License

MIT License