NihonBW 🇯🇵

📜 Abstract

This work focuses on seeing a comparison between ancient and modern approaches to coloring the image where I chose two models which were Pix2Pix and Instance aware Colorization, where a notable difference was found in the coloring of their images.

• keywords: generative adversarial network , deep learning , instance , pix2pix , colorization , gray scale images

The paper is available at this here.

Pipeline

🆕 Update

• 27/02/2021 :

• We already have the first version of our work available here.

• The Streamlit is almost finished ,the Pix2Pix model is loaded and ready to test and the InstColorization model is loaded and ready to test.

• Coming soon.

📖 Content

The following tree shows the structure of the application:

| -master-NihonBW/
|   |-BW/
|       |-Images/
|       |-Result/
|   |-RC/
|       |-Images/
|   |-env/..
|   |-test_pix2pix/..
|   |-imgs/..
|   |-InstColorization/..
|   |-pytorchpix2pix/..
|         |-data/..
|         |-datasets/..
|         |-imgs/..
|         |-options/..
|         |colorization/..
|         |-models/..
|         |..   
|   |-Main.py
|   |-.gitignore
|   |-README.md
|   |-requeriment.txt

Prerequisites

• Linux
• Python 3
• GPU + CUDA CuDNN

Getting Started

Installation

1. Install the envariment

pip install virtualenv

2. Install all dependencies with the command

pip install -r requirements.txt.

3. Clone this repo:

git clone https://github.com/alexliqu09/NihonBW.git

Train models

Pix2Pix

1. If you want to train the model in local , first you need clone the original repository Pix2Pix.
2. Now you should create a dir in pytorch-CycleGAN-and-pix2pix / datasets / with the name colorization and inside of this dir you create train dir , now in train dir put your color images of trains .
3. Finally , you only follow the script of Pix2Pix repository and use the follow command

python train.py --dataroot ./datasets/colorization --name color_pix2pix --model colorization 

4. Search the dir checkpoints/color_pix2pix the weigth latest_net_G.pth .

5. Now in the dir /NihongoBW/pytorchpix2pix/checkpoints/ create the dir experiment_name and you need to move the latest_net_G.pth in the dir /NihongoBW/pytorchpix2pix/checkpoints/experiment_name/.

• Note: If you want to train the model in colab , I used is available here.
• Note: The dataset used me is available here

InstColorization

1. If you want to train the model in local , first you need clone the original repository InstColorization.
2. Now you should create a dir in InstColorization/train_dataset/ with the name colorization and inside of this dir you create train dir , now in train dir put your color images of trains .
3. Finally , you only follow the script of InstColorization repository and use the follow command

#Prepare the data

python inference_bbox.py --test_img_dir ./train_dataset/train --filter_no_obj

#First Module

mkdir ./checkpoints/coco_full

cp ./checkpoints/siggraph_retrained/latest_net_G.pth ./checkpoints/coco_full/

python train.py --stage full --name coco_full --sample_p 1.0 --niter 100 --niter_decay 50 --load_model --lr 0.0005 --model train --fineSize 256 --batch_size 16 --display_ncols 3 --display_freq 1600 --print_freq 1600 --train_img_dir      ./train_dataset/train 

#Second Module

mkdir ./checkpoints/coco_instance
cp ./checkpoints/coco_full/latest_net_G.pth ./checkpoints/coco_instance/
python train.py --stage instance --name coco_instance --sample_p 1.0 --niter 100 --niter_decay 50 --load_model --lr 0.0005 --model train --fineSize 256 --batch_size 16 --display_ncols 3 --display_freq 1600 --print_freq 1600 --train_img_dir ./train_dataset/train 

#Third Module

mkdir ./checkpoints/coco_mask
cp ./checkpoints/coco_full/latest_net_G.pth ./checkpoints/coco_mask/latest_net_GF.pth
cp ./checkpoints/coco_instance/latest_net_G.pth ./checkpoints/coco_mask/latest_net_G.pth
cp ./checkpoints/coco_full/latest_net_G.pth ./checkpoints/coco_mask/latest_net_GComp.pth
python train.py --stage fusion --name coco_mask --sample_p 1.0 --niter 10 --niter_decay 20 --lr 0.00005 --model train --load_model --display_ncols 4 --fineSize 256 --batch_size 1 --display_freq 500 --print_freq 500 --train_img_dir          ./train_dataset/train 

4. Search the dir NihonBW/InstColorization/checkpoints/ the weigth move and change the dir checkpoints/ for the checkpoints/ where you trained.

• Note: If you want to train the model in colab , I used is available here.

Run the work

If You want to proof the work , you need mov to dirs and position yourself with NihonBW/ and run this command

streamlit run main.py

• Note:

Local host : http://localhost:8501 
Network URL:  http://192.168.0.5:8501

below a demonstration of this work.

"NihonBW"

👨🏽‍💻 Maintainer

• Alexander Leonardo Lique Lamas, Github: alexliqu09 Email: alexander.lique.l@uni.pe

🙏🏽 Special thanks

• Many thanks to the repository of junyanz author of Pix2Pix,this work would not have been possible without your repository.
• Many thanks to the repository of ericsujw author of InstColorization,this work would not have been possible without your repository.

Citation

If you find InstColorization useful , please cite this paper:

@inproceedings{Su-CVPR-2020,
  author = {Su, Jheng-Wei and Chu, Hung-Kuo and Huang, Jia-Bin},
  title = {Instance-aware Image Colorization},
  booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2020}
}

If you find Pix2Pix useful , please cite this paper:

@inproceedings{isola2017image,
  title={Image-to-Image Translation with Conditional Adversarial Networks},
  author={Isola, Phillip and Zhu, Jun-Yan and Zhou, Tinghui and Efros, Alexei A},
  booktitle={Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on},
  year={2017}
}