\begin{figure*}\begin{minipage}{0.63\linewidth}\includegraphics[width=1.0\hsize]{HPatches_curve.pdf}\end{minipage}\hfill\begin{minipage}{0.34\linewidth}\tiny\renewcommand\arraystretch{1.12}\resizebox{0.95\textwidth}{!}{\begin{tabular}{llcc}\toprule[0.25mm]\multicolumn{1}{c}{\begin{tabular}[c]{@{}c@{}}Local\\ Methods\end{tabular}}& \multicolumn{1}{c}{Method}& \#Features & \#Matches \\ \hline------& SMAP &6.0K & \textbf{2.8K} \\------& LoFTR &6.4K &2.6K \\------& Sparse-NCNet &2.0K &2.0K \\ \hlineD2-Net & NN &6.0K &2.5K \\R2D2 & NN &5.0K &1.6K \\ASLFeat & NN &4.0K &2.0K \\HAN & NN &3.9K &2.0K \\ \bottomrule[0.25mm]\end{tabular}}\end{minipage}\caption{The mean matching accuracy (MMA) of each method at different matching thresholds (in pixels). Our method achieves the best performance after the threshold of 4px.We report the mean number of detected features and predicted correspondences.}
\end{figure*}