ChromaGAN

Official Keras Implementation of ChromaGAN: Adversarial Picture Colorization with Semantic Class Distribution [WACV 2020] [arXiv] [Supplementary Material]

Network Architecture

Prerequisits

Linux

Python 3

NVIDIA GPU + CUDA CuDNN (CPU mode and CUDA without CuDNN may work with minimal modification, but untested)

Getting Started

Clone Repository

git clone https://github.com/pvitoria/ChromaGAN
cd ChromaGAN/

Requirements

pip install -r requirements.txt

Download the dataset

Download dataset and place it in the /DATASET/ folder. We have train our model with ImageNet dataset. You can download it from here

Network Parameters

All the parameters can be modified from the config.py file. Note: Modify the name of the dataset in the config file in DATASET. For each test you can modify the folder name in TEST_NAME. The variable PRETRAINED should be changed by the name of your pretrained colorization file.

import os

# DIRECTORY INFORMATION
DATASET = "imagenet" # modify
TEST_NAME ="test1" # modify
ROOT_DIR = os.path.abspath('../')
DATA_DIR = os.path.join(ROOT_DIR, 'DATASET/'+DATASET+'/')
OUT_DIR = os.path.join(ROOT_DIR, 'RESULT/'+DATASET+'/')
MODEL_DIR = os.path.join(ROOT_DIR, 'MODEL/'+DATASET+'/')
LOG_DIR = os.path.join(ROOT_DIR, 'LOGS/'+DATASET+'/')

TRAIN_DIR = "train"
TEST_DIR = "test"

# DATA INFORMATION
IMAGE_SIZE = 224
BATCH_SIZE = 10


# TRAINING INFORMATION
PRETRAINED = "my_model_colorization.h5" 
NUM_EPOCHS = 5

Training

To train the network:

cd ChromaGAN/SOURCE/
python ChromaGAN.py

Models are saved to ./MODELS/DATASET/TEST_NAME/

Testing

To test the network you can either run the code directly from Colab using our Demo or run the code as follows :

cd ChromaGAN/SOURCE/
python ChromaGANPrint.py

Images are saved to ./RESULT/DATASET/TEST_NAME/

Pretrained Weights

You can donwload the pretrained weights from here. In order to test the network you should use the file called ` my_model_colorization.h5.

Citation

If you use this code for your research, please cite our paper ChromaGAN: An Adversarial Approach for Picture Colorization:

@inproceedings{vitoria2020chromagan,
  title={ChromaGAN: Adversarial Picture Colorization with Semantic Class Distribution},
  author={Vitoria, Patricia and Raad, Lara and Ballester, Coloma},
  booktitle={The IEEE Winter Conference on Applications of Computer Vision},
  pages={2445--2454},
  year={2020}
}

Aknowledgments

The authors acknowledge partial support by MICINN/FEDER UE project, reference PGC2018-098625-B-I00 VAGS, and by H2020-MSCA-RISE-2017 project, reference 777826 NoMADS. We also thank the support of NVIDIA Corporation for the donation of GPUs used in this work.

Additional Resources

Run inferences using Docker

You can colorize arbitrary grayscale images using a pre-packaged Docker image from the Replicate registry: https://beta.replicate.ai/pvitoria/ChromaGAN The docker has been made by [Andreas Jansson] Andreas Jansson.

Mentions in the Media

ChromaGAN has been recomendeed as one of the "10 Data Science Projects Every Beginner should add to their Portfolio". [Link]

superhighlevel / chromagan Goto Github PK

chromagan's Introduction