Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/fxia22/pixeldtgan
A torch implementation of "Pixel-Level Domain Transfer"
https://github.com/fxia22/pixeldtgan
domain-transfer gan torch
Last synced: 1 day ago
JSON representation
A torch implementation of "Pixel-Level Domain Transfer"
- Host: GitHub
- URL: https://github.com/fxia22/pixeldtgan
- Owner: fxia22
- License: mit
- Created: 2016-07-12T07:33:01.000Z (over 8 years ago)
- Default Branch: master
- Last Pushed: 2017-11-01T18:35:08.000Z (about 7 years ago)
- Last Synced: 2023-10-20T19:55:38.890Z (about 1 year ago)
- Topics: domain-transfer, gan, torch
- Language: Lua
- Homepage: http://arxiv.org/abs/1603.07442
- Size: 2.37 MB
- Stars: 264
- Watchers: 8
- Forks: 46
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Pixel Level Domain Transfer
A torch implementation of "Pixel-Level Domain Transfer". based on [dcgan.torch](https://github.com/soumith/dcgan.torch).
# Dataset
The dataset used is "LookBook", from [Donggeun Yoo](https://dgyoo.github.io).
# Training
To train the model, put the LOOKBOOK dataset under repository, resize images to 64*64. Prepare the dataset using `prepare_data.ipynb`.
Then run
```
th main.lua
```
You can tune the parameters, such as number of filters, optimizer, etc.
# Example results
Example results on LOOKBOOK dataset(top), left is input, right is generated clothes. Results on a similar dataset (bottom).
More results will be added soon.
# Monitor the performance
- Install display package with: `luarocks install https://raw.githubusercontent.com/szym/display/master/display-scm-0.rockspec`
- Start the server with: `th -ldisplay.start`
- Open this URL in your browser: `http://localhost:8000`
Below shows the results after 7 epochs, each 3*1 block is generated cloth, true cloth, input image. Errors of G, D, and A network will be plotted.
