https://github.com/bgavran/gan_lecture_materials
Generative Adversarial Networks presentation and workshop materials
Science Score: 13.0%
This score indicates how likely this project is to be science-related based on various indicators:
-
○CITATION.cff file
-
✓codemeta.json file
Found codemeta.json file -
○.zenodo.json file
-
○DOI references
-
○Academic publication links
-
○Academic email domains
-
○Institutional organization owner
-
○JOSS paper metadata
-
○Scientific vocabulary similarity
Low similarity (11.9%) to scientific vocabulary
Repository
Generative Adversarial Networks presentation and workshop materials
Statistics
- Stars: 7
- Watchers: 2
- Forks: 3
- Open Issues: 1
- Releases: 0
Metadata Files
README.md
Generative Adversarial Networks Lecture
This repository contains the presentation and the code for the workshop for my GAN lecture on which i presented on: * Petnica Summer School of Machine Learning PSIML 2017, 2018. * Ruđer Bošković Institute internal seminars IRB * TakeLab internal seminars TL
Presentation
Different versions of the presentation can be found in multiple branches. Newest version can be found here (July 2018).
Workshop starting point
You will be training your own Wasserstein GAN with Gradient Penalty on the Labeled Faces in the Wild dataset.
The code uses Python 3.6 and TensorFlow 1.2.0rc1 version (but TensorFlow >= 1.0 should be fine).
The data loading pipeline and most of the boilerplate code is already written.
Your task is to download the dataset and fill in the missing parts of the code.
You should be able to train a rudimentary WGAN-GP with fully connected generator and critic on a low resolution Faces dataset in a matter of minutes, even on the CPU of a low-end PC.
Step 1. - Download the dataset
The recommended version is the deep funneled version.
The code will look for images recursively in the code/data/lfw-deepfunnelled directory.
Step 2. - Fill in the missing parts of the code!
Starting point is the main.py file, which sets up and trains your model, but the only file you need to edit is code/src/wgan.py.
There are three things you need to implement: * Cost functions for the critic and the generator * Optimizers for the critic and the generator * Training regime
Network architectures for the generators and the critics are already provided. Place for your code is marked in the file. Although you don't need to touch anything else to make it work, feel free to experiment and adjust the code to your liking!
Step 3. - Train your model!
Run the main.py and check your model performance in TensorBoard!
Run the tensorboard --logdir=log --reload_interval=1 from your code/ directory and access the localhost:6006 address from your browser.
You'll be able to track values of losses, generated images as well as the distribution of weights and gradients of all the parameters in your network.
For reference, Intel i5-4200U CPU @ 1.60GHz trains the basic shape of the face within first 200 iterations!
Step 4.
???
Step 5.
Profit!!!
Feel free to expriment with various architectures for generator and the critic!
Download some other dataset!
Try removing the Lipschitz regularization and see how the model fails!
Try to figure out your own way to enforce the Lipschitz constraint! (and publish a paper about it!)
If you have access to a GPU farm, try generating HD images!
Owner
- Name: Bruno Gavranović
- Login: bgavran
- Kind: user
- Location: London, United Kingdom
- Company: Symbolica
- Website: www.brunogavranovic.com
- Repositories: 26
- Profile: https://github.com/bgavran
Principal Scientist - Categorical Deep Learning @symbolica-ai
GitHub Events
Total
Last Year
Issues and Pull Requests
Last synced: over 1 year ago
All Time
- Total issues: 1
- Total pull requests: 0
- Average time to close issues: N/A
- Average time to close pull requests: N/A
- Total issue authors: 1
- Total pull request authors: 0
- Average comments per issue: 0.0
- Average comments per pull request: 0
- Merged pull requests: 0
- Bot issues: 0
- Bot pull requests: 0
Past Year
- Issues: 0
- Pull requests: 0
- Average time to close issues: N/A
- Average time to close pull requests: N/A
- Issue authors: 0
- Pull request authors: 0
- Average comments per issue: 0
- Average comments per pull request: 0
- Merged pull requests: 0
- Bot issues: 0
- Bot pull requests: 0
Top Authors
Issue Authors
- bgavran (1)