Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/gazeux33/neuralstyletransfert
Implementation of Neural Style Transfert using PyTorch
https://github.com/gazeux33/neuralstyletransfert
Last synced: about 1 month ago
JSON representation
Implementation of Neural Style Transfert using PyTorch
- Host: GitHub
- URL: https://github.com/gazeux33/neuralstyletransfert
- Owner: Gazeux33
- Created: 2024-05-04T18:39:41.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2024-05-21T07:07:56.000Z (8 months ago)
- Last Synced: 2024-05-21T21:31:23.951Z (8 months ago)
- Language: Python
- Size: 16 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Neural Style Transfert(NST)
This project is an implementation of Neural Style Transfer using PyTorch. Neural Style Transfer is a deep learning technique that merges the style of one image with the content of another, creating visually striking results by blending artistic features with photographic details.
## What is NST ?
## Technical specifications
| Property | Value |
|----------------|---------------|
| Framework | PyTorch |
| Device | MAC M2 |
| Optimizer | LBFGS |
| Time for 1 image | ~20 min |
## How does it work ?
### Extract characteristics with VGG19
The VGG19 pre-trained model is a convolutional neural network with 19 layers, including 16 convolutional layers and 3 fully connected layers, featuring 3x3 convolutional filters and max pooling, totaling approximately 143.67 million parameters for image classification tasks.
we start by extracting the characteristics of content and style
---
### Optimization Loop
The ititial image can be:
- `content`
- `style`
- `random noise`
---
### Loss Function
$$
L_{\text{total}} = \alpha L_{\text{content}} + \beta L_{\text{style}} + \gamma L_{\text{TV}}
$$
- `L_Content` is the content loss.
- `L_Style` is the style loss.
- `L_TV` is the total variation loss.
- `alpha`, `beta`, and `gamma` are hyperparamters for each loss.
---
### Style Loss
---
### Content Loss
