https://github.com/aazuspan/mnist-polar
An experiment classifying MNIST digits in 1D with polar aggregation
https://github.com/aazuspan/mnist-polar
1d-cnn mnist polar-coordinates pytorch
Last synced: 3 months ago
JSON representation
An experiment classifying MNIST digits in 1D with polar aggregation
- Host: GitHub
- URL: https://github.com/aazuspan/mnist-polar
- Owner: aazuspan
- Created: 2025-02-17T02:12:23.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2025-03-05T01:22:46.000Z (3 months ago)
- Last Synced: 2025-03-05T02:25:13.807Z (3 months ago)
- Topics: 1d-cnn, mnist, polar-coordinates, pytorch
- Language: Jupyter Notebook
- Homepage:
- Size: 12.5 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
An experiment classifying MNIST digits in 1D with polar aggregation.
See the [blog post](http://aazuspan.dev/blog/classifying-mnist-as-1d-signals/) for details.
## Method
1. Digit images are converted to 1D signals by [aggregating over the radial dimension](https://github.com/aazuspan/mnist-polar/blob/340c493b35e4b03f896088350ab06587db6fdc46/src/polar_mnist/transforms.py#L8-L29) in polar space.
https://github.com/user-attachments/assets/6833013f-0c52-4c75-9b5a-0e4b607e8280
2. Signals are classified using a [1D CNN](https://github.com/aazuspan/mnist-polar/blob/main/src/polar_mnist/model.py) with circular padding.
