Xoryn-II / source /README.md
PureOne's picture
Release Xoryn II finite-grid image recovery benchmark
8f0b027 verified
|
Raw History Blame Contribute Delete
3.06 kB
# Xoryn II / A-47 — The finite-grid recovery theorem
**Statement, Master.** This archive contains the proofs and the exact inputs for the finite-grid sensing and recognition tests. The sensing model uses independently and uniformly shifted periodic images. The compressed readouts are simulated from the recorded image arrays.
## Contents
| File | Contents |
| --- | --- |
| Xoryn_II_Finite_Grid_Recovery.pdf | Complete manuscript with proofs, figures, and comparison |
| Xoryn_II.tex | Manuscript source |
| grid.py | Fourier sensing, selected cubic inverse, and exact graph check |
| verify.py | Reproducible population, spectral, and noisy image tests |
| benchmark_inputs.npz | Frozen digits and RGB inputs with source metadata |
| results.json | Machine-readable numerical results |
| results_table.tex, recognition.pdf, rgb.pdf | Generated report table and figures |
| requirements.txt | Tested Python dependencies |
| SHA256SUMS.txt | Checksums for the individual files |
## Reproduce
Use Python 3.11 or later. The reported run used Python 3.12.14.
```sh
python -m pip install -r requirements.txt
python verify.py --output reproduced
```
The digit orientation test and the numerical checks are seeded. Data are bundled, so verification needs no dataset downloads. A full run checks all 1,797 dataset entries for spectral support and reconstructs them from exact averaging over all shifts. It uses one sine measurement on full support and two on supports with holes. Its noisy tests are balanced known-template decisions. The RGB reconstruction is a noiseless population test. Rectangular views are treated periodically.
Run `pdflatex Xoryn_II.tex` twice from the extracted directory to rebuild the supplied manuscript. Its figures and table are included.
## Data attribution
- The digit arrays are the 1,797 examples distributed with [scikit-learn's load_digits](https://scikit-learn.org/stable/modules/generated/sklearn.datasets.load_digits.html). Original dataset by Ethem Alpaydın and Fevzi Kaynak, [Optical Recognition of Handwritten Digits](https://doi.org/10.24432/C50P49), UCI Machine Learning Repository, [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). The figure shows selected points and their periodic inversions.
- RGB data consist of 12 cropped patches from the two images distributed with [scikit-learn's load_sample_images](https://scikit-learn.org/stable/modules/generated/sklearn.datasets.load_sample_images.html), licensed under [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/). [China photo by Daniel Büchele](https://www.flickr.com/photos/danielbuechele/6061409035) and [flower photo source at vultilion](https://www.flickr.com/photos/vultilion/6056698931). The crop origins and image names are in the archive. The manuscript shows crops and their reconstructions.
The initial Xoryn theorem was developed from `frac(4).txt`, whose SHA-256 was `f0ed4fc754dae89fd4d1d7fb146f94bb8f996c74e82816e04de09b521ddc1e04`. This sequel's proofs and benchmark start from the model stated in the manuscript.