https://github.com/fork123aniket/cyclegan-for-image-translation
Implementation of Cycle-consistent Generative Adversarial Networks for Image-to-Image Translation in Keras
https://github.com/fork123aniket/cyclegan-for-image-translation
convolutional-neural-networks cycle-gan generative-adversarial-networks image-to-image-translation keras keras-models keras-neural-networks
Last synced: 7 months ago
JSON representation
Implementation of Cycle-consistent Generative Adversarial Networks for Image-to-Image Translation in Keras
- Host: GitHub
- URL: https://github.com/fork123aniket/cyclegan-for-image-translation
- Owner: fork123aniket
- License: mit
- Created: 2022-05-20T08:18:15.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2023-05-01T12:00:53.000Z (over 2 years ago)
- Last Synced: 2025-01-16T06:25:23.588Z (9 months ago)
- Topics: convolutional-neural-networks, cycle-gan, generative-adversarial-networks, image-to-image-translation, keras, keras-models, keras-neural-networks
- Language: Python
- Homepage:
- Size: 3.67 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# CycleGAN for Image Translation
This repository aims at developing a ***CycleGAN*** model to translate photos of horses to zebras and vice versa. Here, it is not necessary to provide paired examples in the dataset to ***CycleGAN***, i.e. it does not require examples of photographs before and after the translation in order to train the model. Instead, the model is able to use a collection of photographs from each domain and extract and harness the underlying style of images in the collection in order to perform the translation.
## Requirements
- `numpy`
- `matplotlib`
- `keras`
## Usage
### Data
- The ***horse2zebra*** dataset can be downloaded from the [***CycleGAN webpage***](https://people.eecs.berkeley.edu/~taesung_park/CycleGAN/datasets/horse2zebra.zip).
- In the dataset, the ***A*** category refers to horse and ***B*** category refers to zebra, and the dataset is comprised of train and test elements. However, the current implementation loads all photographs and use them as a training dataset.
- All photographs are of the same size, i.e. *256 * 256*.
- To prepare the image dataset, considered here, in order to properly train the ***CycleGAN*** model, run `Data.py`
### Training and Testing
- To train and test the ***CycleGAN*** model along with plotting the results, run `Cycle_GAN_to_translate_Images.py`
- To translate multiple images using ad-hoc image translation process with the help of trained ***CycleGAN***, run `Image_Translation_with_Cycle_GAN_Generators.py`
- To translate merely individual images using the trained ***CycleGAN*** model, run `Single_Image_Translation_with_Cycle_GAN_Generators.py`
- All hyperparameters to control training and testing of the ***CycleGAN*** model in single as well as multiple image translation settings are provided in their respective `.py` files.
- The average loss for both generator and discriminator components of the ***CycleGAN*** model for each of the two catagories (***Horse (A)*** and ***Zebra (B)***) are printed after every epoch.
## Output Samples
### Few Examples of Image Dataset
### Image-to-Image Translation Results
| Horse-to-Zebra Results | Zebra-to-Horse Results |
| ------------------------- |:----------------------------:|
|  |  |
### Multiple Images Translation
| Horse-to-Zebra Results | Zebra-to-Horse Results |
| ------------------------- |:----------------------------:|
|  |  |
### Single Image Translation
| Real Horse Image | Generated Zebra Image |
| ------------------------- |:----------------------------:|
|  |  |