sanjay035 / Sketch2Color-anime-translation

Given a simple anime line-art sketch the model outputs a decent colored anime image using Conditional-Generative Adversarial Networks (C-GANs) concept.

Sketch2Color-anime-translation:

Given a simple line-art anime sketch the model should output a decent colored anime image using Conditional-Generative Adversarial Networks (C-GANs).

Implemented the anime sketch colorization using the reference(s) paper-1 below.

• Dataset: Anime sketch colorization pair
• Blog on: Towards Data Science

References:

[1] https://arxiv.org/abs/1705.01908

[2] https://machinelearningmastery.com

[3] https://github.com/soumith/ganhacks

Prerequisites:

• Miniconda
• Tensorflow 1.15
• NVIDIA GPU (8G memory) + CUDA cuDNN

Installation:

• Clone this repo:

> git clone https://github.com/sanjay235/Sketch2Color-anime-translation.git
> cd Sketch2Color-anime-translation

• Install Miniconda from https://docs.conda.io/en/latest/miniconda.html
• Install Tensorflow and dependencies from https://www.tensorflow.org/install/
• Open the anaconda prompt and create a new environment with environment.yml file, the first line of the yml file sets the new environment's name:

> conda env create -f environment.yml

• Then activate the environment created and launch the jupyter:

> conda activate Sketch2ColorAnimeTranslation
> jupyter notebook

Dataset:

• Download and extract the dataset to current directory.
• Open an another anaconda prompt with same environment activated and run the pre-processing script as below,

> python DataPreprocessing.py

Training and testing:

• For the training process refer the Sketch2Color_Anime.ipynb notebook.
• For the evaluation on test data refer the final.ipynb notebook.

C-GAN Architecture:

Tensorboard logs:

In progress training results:

At epoch 30,

At epoch 40,

At epoch 43,

Prediction on test sketches:

Contribute:

Correction & Contribution are always welcome!! 😃