Datasets:
Formats:
parquet
Size:
1K - 10K
Tags:
computer-vision
orbit-recovery
projected-mra
periodic-translation
image-reconstruction
bispectrum
License:
File size: 1,366 Bytes
8f0b027 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 | \begin{tabularx}{\linewidth}{@{}Xr@{}}
\toprule Result & Measurement\\ \midrule
Fully supported recorded digits, out of 1,797 & 1,678\\
Largest noiseless relative translation-orbit error on these digits & $2.57\times10^{-15}$\\
Digit readings per view, one-tap versus full & 35 versus 64\\
Remaining digits with spectral holes, recovered with two taps & 119\\
Largest noiseless relative error on those digits & $2.16\times10^{-14}$\\
Natural-image RGB crops reconstructed, out of 12 & 12\\
Largest noiseless relative error on RGB crops, excluding DC & $1.97\times10^{-14}$\\
RGB readings per view, two-tap versus full & 1,544 versus 3,072\\
Two-tap transfer across 57 synthetic cases & 17,686 cubic coordinate checks\\
\bottomrule\end{tabularx}
\medskip\noindent\textbf{Noisy point-inversion decisions.} Each setting uses 1,024 balanced trials across 512 known digit templates.
{\small\begin{tabularx}{\linewidth}{@{}rcc>{\centering\arraybackslash}Xr@{}}
\toprule Views & Noise SD & One fixed tap & Full-view cubic & Folded Bayes\\ \midrule
32 & 0.5 & 98.9\% & 99.8\% & 50.0\%\\
32 & 1 & 95.7\% & 98.4\% & 50.0\%\\
32 & 2 & 74.7\% & 88.7\% & 50.0\%\\
32 & 4 & 56.6\% & 58.7\% & 50.0\%\\
128 & 0.5 & 99.3\% & 99.6\% & 50.0\%\\
128 & 1 & 98.8\% & 99.6\% & 50.0\%\\
128 & 2 & 87.7\% & 96.9\% & 50.0\%\\
128 & 4 & 61.2\% & 70.9\% & 50.0\%\\
\bottomrule\end{tabularx}}
|