Czytnik Polskich Banknotów is an Open Source mobile app, which goal is to help visually handicapped people live the life we all do. By detecting our local currency's denomination (PLN - Polish zloty) using neural networks, it allows a user to know what banknote they're holding. The application provides support on a daily basis, such as going shopping, using ATMs, buying tickets or visiting restaurants.
The app is currently maintained by jankejc:
- GitHub
- email: [email protected]
Feel free to add something to the project. At that moment we are looking especially for iOS, Android, AI model contributors. We highly appreciate feedback from those who would like to test the app. We also want to reach out to those who have experience in generating data for model training purpose. We are talking about i.e. script in Blender that will generate a lot of good data for any currency in the world.
Thanks to this we can collect new features that may transform this app to be even more useful for those who live slightly different life.
Available at Play Store
Not currently aviable at App Store.
Short description of technologies used in this project.
We're developing two versions of this application - one for Android and one for iOS. Both of them are written natively using corresponding technologies - Kotlin for Android and Swift for iOS. Currently, supported system versions are:
- iOS 11.0+ (iPhone 5s+)
- Android 7+
The neural network we're working on is based on PyTorch Image classification which is then exported into TensorFlow Lite. We used YOLOv5s-cls architecture to train our best model - created by Team Android.
We're using datasets created by previous teams working on this subject (information about those are given in Acknowledgments section). About numbers, we currently have around 3500-4800 images per denomination, mostly in good quality with 0 or 1 banknote at it.
For help, please contact authors of this project or create an issue on GitHub.
- Jacek Ardanowski
- Bartłomiej Gawrych
- Adam Grabowski
- Kamil Jabłoński
- Bartosz Kuncer
- Aleksander Madajczak
- Kamil Pokornicki
- Karol Dziki
We were using datasets created by previous teams working on this subject.
Polish banknotes / Polskie banknoty
Thanks to kindness of Ph.D. Jacek Rumiński from Biomedical Engineering Department, we were able to use DGX Station, which is a powerful machine learning server, to train our neural network.
This work has been partially supported by Statutory Funds of Electronics, Telecommunications and Informatics Faculty, Gdańsk University of Technology.
This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent