All Projects → ImagingLab → Colorizing With Gans

ImagingLab / Colorizing With Gans

Licence: apache-2.0
Grayscale Image Colorization with Generative Adversarial Networks. https://arxiv.org/abs/1803.05400

Programming Languages

python
139335 projects - #7 most used programming language

Projects that are alternatives of or similar to Colorizing With Gans

Iseebetter
iSeeBetter: Spatio-Temporal Video Super Resolution using Recurrent-Generative Back-Projection Networks | Python3 | PyTorch | GANs | CNNs | ResNets | RNNs | Published in Springer Journal of Computational Visual Media, September 2020, Tsinghua University Press
Stars: ✭ 202 (-3.35%)
Mutual labels:  generative-adversarial-network, gans
3d Recgan Extended
🔥3D-RecGAN++ in Tensorflow (TPAMI 2018)
Stars: ✭ 98 (-53.11%)
Mutual labels:  generative-adversarial-network, gans
Bicyclegan
Toward Multimodal Image-to-Image Translation
Stars: ✭ 1,215 (+481.34%)
Mutual labels:  generative-adversarial-network, gans
Co Mod Gan
[ICLR 2021, Spotlight] Large Scale Image Completion via Co-Modulated Generative Adversarial Networks
Stars: ✭ 46 (-77.99%)
Mutual labels:  generative-adversarial-network, gans
Gesturegan
[ACM MM 2018 Oral] GestureGAN for Hand Gesture-to-Gesture Translation in the Wild
Stars: ✭ 136 (-34.93%)
Mutual labels:  generative-adversarial-network, gans
Pacgan
[NeurIPS 2018] [JSAIT] PacGAN: The power of two samples in generative adversarial networks
Stars: ✭ 67 (-67.94%)
Mutual labels:  generative-adversarial-network, gans
Tagan
An official PyTorch implementation of the paper "Text-Adaptive Generative Adversarial Networks: Manipulating Images with Natural Language", NeurIPS 2018
Stars: ✭ 97 (-53.59%)
Mutual labels:  generative-adversarial-network, gans
Data Efficient Gans
[NeurIPS 2020] Differentiable Augmentation for Data-Efficient GAN Training
Stars: ✭ 682 (+226.32%)
Mutual labels:  generative-adversarial-network, gans
Ganspapercollection
Stars: ✭ 130 (-37.8%)
Mutual labels:  generative-adversarial-network, gans
Cyclegan
Software that can generate photos from paintings, turn horses into zebras, perform style transfer, and more.
Stars: ✭ 10,933 (+5131.1%)
Mutual labels:  generative-adversarial-network, gans
Bringing Old Photos Back To Life
Bringing Old Photo Back to Life (CVPR 2020 oral)
Stars: ✭ 9,525 (+4457.42%)
Mutual labels:  generative-adversarial-network, gans
A Pytorch Tutorial To Super Resolution
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network | a PyTorch Tutorial to Super-Resolution
Stars: ✭ 157 (-24.88%)
Mutual labels:  generative-adversarial-network, gans
Contrastive Unpaired Translation
Contrastive unpaired image-to-image translation, faster and lighter training than cyclegan (ECCV 2020, in PyTorch)
Stars: ✭ 822 (+293.3%)
Mutual labels:  generative-adversarial-network, gans
Cnn Paper2
🎨 🎨 深度学习 卷积神经网络教程 :图像识别,目标检测,语义分割,实例分割,人脸识别,神经风格转换,GAN等🎨🎨 https://dataxujing.github.io/CNN-paper2/
Stars: ✭ 77 (-63.16%)
Mutual labels:  gans, convolutional-networks
Gans In Action
Companion repository to GANs in Action: Deep learning with Generative Adversarial Networks
Stars: ✭ 748 (+257.89%)
Mutual labels:  generative-adversarial-network, gans
Doppelganger
[IMC 2020 (Best Paper Finalist)] Using GANs for Sharing Networked Time Series Data: Challenges, Initial Promise, and Open Questions
Stars: ✭ 97 (-53.59%)
Mutual labels:  generative-adversarial-network, gans
Fast Srgan
A Fast Deep Learning Model to Upsample Low Resolution Videos to High Resolution at 30fps
Stars: ✭ 417 (+99.52%)
Mutual labels:  generative-adversarial-network, gans
Von
[NeurIPS 2018] Visual Object Networks: Image Generation with Disentangled 3D Representation.
Stars: ✭ 497 (+137.8%)
Mutual labels:  generative-adversarial-network, gans
3d Recgan
🔥3D-RecGAN in Tensorflow (ICCV Workshops 2017)
Stars: ✭ 116 (-44.5%)
Mutual labels:  generative-adversarial-network, gans
Generative adversarial networks 101
Keras implementations of Generative Adversarial Networks. GANs, DCGAN, CGAN, CCGAN, WGAN and LSGAN models with MNIST and CIFAR-10 datasets.
Stars: ✭ 138 (-33.97%)
Mutual labels:  generative-adversarial-network, gans

Image Colorization with Generative Adversarial Networks

In this work, we generalize the colorization procedure using a conditional Deep Convolutional Generative Adversarial Network (DCGAN) as as suggested by Pix2Pix. The network is trained on the datasets CIFAR-10 and Places365. Some of the results from Places365 dataset are shown here.

Prerequisites

  • Linux
  • Tensorflow 1.7
  • NVIDIA GPU (12G or 24G memory) + CUDA cuDNN

Getting Started

Installation

  • Clone this repo:
git clone https://github.com/ImagingLab/Colorizing-with-GANs.git
cd Colorizing-with-GANs
pip install -r requirements.txt

Dataset

  • We use CIFAR-10 and Places365 datasets. To train a model on the full dataset, download datasets from official websites. After downloading, put then under the datasets folder.

Training

  • To train the model, run main.py script
python main.py
  • To train the model on places365 dataset with tuned hyperparameters:
python train.py \
  --seed 100 \
  --dataset places365 \
  --dataset-path ./dataset/places365 \
  --checkpoints-path ./checkpoints \
  --batch-size 16 \
  --epochs 10 \
  --lr 3e-4 \
  --label-smoothing 1
  
  • To train the model of cifar10 dataset with tuned hyperparameters:
python train.py \
  --seed 100 \
  --dataset cifar10 \
  --dataset-path ./dataset/cifar10 \
  --checkpoints-path ./checkpoints \
  --batch-size 128 \
  --epochs 200 \
  --lr 3e-4 \
  --lr-decay-steps 1e4 \
  --augment True
  

Test

  • Download the pre-trained weights from here. and copy them in the checkpoints folder.
  • To test the model on a custom image(s), run test.py script:
python test.py \
  --checkpoints-path ./checkpoints \        # checkpoints path
  --test-input ./checkpoints/test \         # test image(s) path
  --test-output ./checkpoints/output \      # output image(s) path

Visual Turing Test

  • Download the pre-trained weights from here. and copy them in the checkpoints folder.
  • To evaluate the model qualitatively using visual Turing test, run test-turing.py:
python test-turing.py
  • To apply time-based visual Turing test run (2 seconds decision time):
python test-turing.py --test-delay 2

Networks Architecture

The architecture of generator is inspired by U-Net: The architecture of the model is symmetric, with n encoding units and n decoding units. The contracting path consists of 4x4 convolution layers with stride 2 for downsampling, each followed by batch normalization and Leaky-ReLU activation function with the slope of 0.2. The number of channels are doubled after each step. Each unit in the expansive path consists of a 4x4 transposed convolutional layer with stride 2 for upsampling, concatenation with the activation map of the mirroring layer in the contracting path, followed by batch normalization and ReLU activation function. The last layer of the network is a 1x1 convolution which is equivalent to cross-channel parametric pooling layer. We use tanh function for the last layer.

For discriminator, we use patch-gan architecture with contractive path similar to the baselines: a series of 4x4 convolutional layers with stride 2 with the number of channels being doubled after each downsampling. All convolution layers are followed by batch normalization, leaky ReLU activation with slope 0.2. After the last layer, a sigmoid function is applied to return probability values of 70x70 patches of the input being real or fake. We take the average of the probabilities as the network output!

Places365 Results

Colorization results with Places365. (a) Grayscale. (b) Original Image. (c) Colorized with GAN.

Citation

If you use this code for your research, please cite our paper Image Colorization with Generative Adversarial Networks:

@inproceedings{nazeri2018image,
  title={Image Colorization Using Generative Adversarial Networks},
  author={Nazeri, Kamyar and Ng, Eric and Ebrahimi, Mehran},
  booktitle={International Conference on Articulated Motion and Deformable Objects},
  pages={85--94},
  year={2018},
  organization={Springer}
}
Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].