SimpleRegion
 |
|---|
| From up to down: real images, soft regions, hard regions, hard regions with random colors |
A simple neural network model for extracting regions from an image.
The result of the model is similar to clustering the pixels of a single image or do a superpixel task.
Principle
-
Neural network classifies image pixels to multiple regions.
-
Average the pixels of the original image corresponding to a region to get the average color of the region.
-
Assign the average color to the corresponding position in the region.
-
During the training process, the classification result is the probability obtained by softmax, not just 0 and 1. This kind of gradient result enables the colors to blend well, thereby generating a soft colored result.
-
Replace the maximum value in the classification probability with 1, and replace the remaining values with 0, and then perform coloring to obtain a hard colored image. This looks like clustering an image or generating superpixel regions.
Training
Run the command line:
python train.py <YOUR DATASET DIRECTORY>Use
If you modify this model, it should be used to separate the edges, solid colors, gradients and lighting of an image.
This is just a simple neural network model. If you have a very complicated idea about extracting regions from an image, I recommend you to use the dataset and model of lllyasviel/DanbooRegion.
Contact
QQ Group: 1044867291
Discord Channel: https://discord.gg/YwWcAS47qb
Citation
@misc{Wu_Simple_Region_2021,
author = {Wu, Hecong},
title = {A Simple Neural Network Model For Extracting Regions From An Image},
month = {10},
year = {2021},
publisher = {GitHub},
journal = {GitHub repository},
url = {https://github.com/HighCWu/SimpleRegion},
}