Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/meetptl04/imagestyletranfer
This project demonstrates how to apply neural style transfer using TensorFlow and a pre-trained VGG19 model. The goal is to combine the content of one image with the style of another to create a stylized image. The repository includes code for image processing, loss computation, style transfer, and visualization
https://github.com/meetptl04/imagestyletranfer
image-processing jupyter-notebook keras matplotlib tenserflow
Last synced: 4 days ago
JSON representation
This project demonstrates how to apply neural style transfer using TensorFlow and a pre-trained VGG19 model. The goal is to combine the content of one image with the style of another to create a stylized image. The repository includes code for image processing, loss computation, style transfer, and visualization
- Host: GitHub
- URL: https://github.com/meetptl04/imagestyletranfer
- Owner: meetptl04
- License: mit
- Created: 2024-09-03T15:28:25.000Z (2 months ago)
- Default Branch: main
- Last Pushed: 2024-11-04T05:19:15.000Z (4 days ago)
- Last Synced: 2024-11-04T06:20:49.587Z (4 days ago)
- Topics: image-processing, jupyter-notebook, keras, matplotlib, tenserflow
- Language: Jupyter Notebook
- Homepage:
- Size: 24.8 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# ImageStyleTranfer
---
# Image Style Transfer
This project demonstrates how to apply neural style transfer using TensorFlow and a pre-trained VGG19 model. The goal is to combine the content of one image with the style of another to create a stylized image. The repository includes code for image processing, loss computation, style transfer, and visualization.
## Overview
The project consists of several components:
1. **Image Processing**: Methods to load, process, and de-process images for neural network input and visualization.
2. **Loss Functions**: Functions to compute content and style losses, including the gram matrix for style loss.
3. **VGG Model**: A class for initializing the VGG19 model and extracting content and style features.
4. **Style Transfer**: Core class for performing the style transfer, including loss computation and gradient updates.
5. **Visualization**: Methods for displaying images using Matplotlib.
## Table of Contents
- [Installation](#installation)
- [Usage](#usage)
- [Classes](#classes)
- [ImageProcessing](#imageprocessing)
- [LossFunctions](#lossfunctions)
- [VGGModel](#vggmodel)
- [StyleTransfer](#styletransfer)
- [Visualization](#visualization)
- [Results](#results)
- [License](#license)
## Installation
1. Clone the repository:
```bash
git clone https://github.com/meetptl04/ImageStyleTranfer.git
cd ImageStyleTranfer
```
2. Install the required libraries:
```bash
pip install tensorflow matplotlib numpy
```
## Usage
1. **Initialize Classes**: Set up the VGG model and style transfer instance.
2. **Run Style Transfer**: Perform the style transfer by specifying paths to the content and style images.
3. **Visualize Results**: Display the original images and the final stylized image.
Example:
```python
# Initialize the classes
vgg_model = VGGModel()
style_transfer = StyleTransfer(vgg_model)
visualizer = Visualization()
# Paths to images
original_image = '/content/the-starry-night.jpg'
style_image = '/content/Fantasy-Garden.png'
# Run the style transfer
best_img, best_loss = style_transfer.run_style_transfer(original_image, style_image)
# Display images
visualizer.show_image_matplotlib(original_image)
visualizer.show_image_matplotlib(style_image)
# Display the resulting image
visualizer.show_results(best_img)
```
## Classes
### ImageProcessing
The `ImageProcessing` class provides static methods for loading and processing images for neural network input. It resizes and normalizes images for VGG19 and performs de-normalization for display.
```python
class ImageProcessing:
@staticmethod
def load_and_process_img(path_to_img):
# Load, resize, and preprocess image
...
@staticmethod
def deprocess_img(processed_img):
# Convert processed image back to viewable format
...
```
### LossFunctions
The `LossFunctions` class defines methods to compute content loss and style loss, including the gram matrix calculation for style loss.
```python
class LossFunctions:
@staticmethod
def get_content_loss(base_content, target):
# Compute content loss
...
@staticmethod
def gram_matrix(input_tensor):
# Calculate gram matrix
...
@staticmethod
def get_style_loss(base_style, gram_target):
# Compute style loss
...
```
### VGGModel
The `VGGModel` class initializes a VGG19 model pre-trained on ImageNet and extracts feature representations for content and style images.
```python
class VGGModel:
def __init__(self):
# Initialize VGG19 model
...
def get_feature_representations(self, content_path, style_path):
# Extract content and style features
...
```
### StyleTransfer
The `StyleTransfer` class applies neural style transfer, computing losses, gradients, and optimizing the image to blend the style of one image with the content of another.
```python
class StyleTransfer:
def __init__(self, vgg_model, style_weight=1e-2, content_weight=1e3):
# Initialize style transfer parameters
...
def compute_loss(self, init_image, gram_style_features, content_features):
# Compute style and content losses
...
def compute_grads(self, init_image, gram_style_features, content_features):
# Compute gradients
...
def run_style_transfer(self, content_path, style_path, num_iterations=1000):
# Perform style transfer and optimize image
...
```
### Visualization
The `Visualization` class provides methods to display images using Matplotlib.
```python
class Visualization:
@staticmethod
def show_results(best_img):
# Display final stylized image
...
@staticmethod
def show_image_matplotlib(image_path):
# Display image from file path
...
```
## Results
Example output of style transfer, showing the original images and the final stylized image.
## License
This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.
---