https://github.com/rapidrabbit76/sketchcolorization

line drawing colorization using pytorch
https://github.com/rapidrabbit76/sketchcolorization

deep-learning gans generative-adversarial-network machine-learning onnx paintschainer pytorch

Last synced: 12 months ago
JSON representation

line drawing colorization using pytorch

• Host: GitHub
• URL: https://github.com/rapidrabbit76/sketchcolorization
• Owner: rapidrabbit76
• License: mit
• Created: 2020-12-11T08:55:26.000Z (over 5 years ago)
• Default Branch: main
• Last Pushed: 2022-09-18T11:06:31.000Z (almost 4 years ago)
• Last Synced: 2023-03-11T11:56:38.047Z (over 3 years ago)
• Topics: deep-learning, gans, generative-adversarial-network, machine-learning, onnx, paintschainer, pytorch
• Language: Python
• Homepage: https://omnissiah.ys2lee.com/
• Size: 8.95 MB
• Stars: 13
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# SketchColorization ([Web](https://omnissiah.ys2lee.com/))

[![web](./src/web_01.png)](https://omnissiah.ys2lee.com/)

![01](./src/01.png)

# Model Structure 

![02](./src/02.jpg)

# Samples

![03](./src/03.jpg)

![04](./src/04.jpg)

# GUI

---

![5](./src/06.png)

# Requirements

- torch==1.7.1

- torchvision==0.82

- numpy==1.19.1

- tensorboard==2.3.0

- tqdm==4.28.1

- opencv_python==4.4.0.46

- scipy==1.5.2

- Pillow==7.2.0

- scikit-learn==0.23.2

- fbs==0.9.0

- onnx==1.7.0

- onnxruntime==1.5.1

- PyQt5==5.15.1

- QDarkStyle==2.8.1

# Dataset 

- We crawled over 700,000 illustrations from [shuushuu-image-board](https://e-shuushuu.net/) and used them for learning.

- We have filtered out noise such as extreme aspect ratio, black and white image, low / high key images and etc.

# Training

- The learning sequence is 1. autoencoder, 2. draft, 3. colorization.

- set hyperparameters.yml, e.g. paths (image_path and line_path, logdir)

- Start learning after adjusting hyperparameters for each learning step

    - run 'python main.py  -M {autoencoder | draft | colorization}'

# Run APP with source code

- download pretrained onnx model [SketchColorizationModel.onnx](https://github.com/rapidrabbit76/SketchColorization/releases)

- Copy model to "app/src/main/resources/base/SketchColorizationModel.onnx"

- cd app 

- fbs run