Need help with Unsupervised-Segmentation?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

130 Stars 30 Forks 15 Commits 2 Opened issues


A high performance impermentation of Unsupervised Image Segmentation by Backpropagation - Asako Kanezaki

Services available


Need anything else?

Contributors list

# 82,786
15 commits


An implementation of Unsupervised Image Segmentation by Backpropagation - Asako Kanezaki 金崎朝子 (東京大学)ICASSP. 2018.

Faster and more elegant than origin version. Speed up, 30s(origin) --> 5s(modify)


Original version Github:

An Interpretation of this algorithm: (Warning: Simplified Chinese)


Necessary: Python 3, Torch 0.4

Unnecessary: skimage, opencv-python(cv2)

Getting Started

Try the high performance code written by me. ``` python3

class Args(object): # You can change the inputimagepath ↓ inputimagepath = 'image/woof.jpg' # image/coral.jpg image/tiger.jpg ```

Or you want to try the code written by the original author.

python3 --input image/woof.jpg

Run this demo, and press WASDQE on the keyboard to adjust the parameters. The image show in the GUI, and the parameters show in terminal in real time. You could choose Algorithm felz or Algorithm slic by commenting the code. * W,S --> parameter 1 * A,D --> parameter 2 * Q,E --> parameter 3



The iterative process: Save the result when the iter_number == 1,2,4,8,16,32,64,128.

The different result of Algorithm felz or Algorithm slic with different parameters.

The left picture: compactness = 10000

The right picture: compactness = 1000

The left picture: Algorithm slic

The right picture: Algorithm felz

Translate 翻译

If you can understand English, then I know you can understand this line of words (and you see this line on GitHub.)


An implementation of Unsupervised Image Segmentation by Backpropagation


In my opinion, this algorithm is well suited for unsupervised segmentation of satellite images, because satellite images have no directionality. It is suitable for this algorithm with a priori assumption. (Priori Assumptions: In general, the regions with the same semantic information on the satellite images tend to occurs in a continuous area)

这个算法很适合做 卫星图片的无监督语义分割任务,因为卫星地图没有方向性,并且地图上带有相同语义信息的区域往往是出现在一起的(符合先验假设)。很适合这种带有这种的先验假设算法。

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.