Dataset Viewer
id
int64 1
213k
| paper_arxiv_id
stringlengths 10
13
| path
float64 | caption
stringlengths 9
127k
⌀ | label
stringlengths 7
125
⌀ | table_text
stringlengths 33
131k
⌀ |
|---|---|---|---|---|---|
1
|
2008.03299
| null | null |
\label{table:ExampleComplex}
|
\begin{tabular}{|c|c|c|c|}
\hline
$p$ & $\dim Z_p$ & $\dim B_p$ & $\beta_p$ \\
\hline
$0$ & $5$ & $3$ & $2$ \\
$1$ & $2$ & $1$ & $1$ \\
$2$ & $0$ & $0$ & $0$ \\
\hline
\end{tabular}
|
2
|
2008.03300
| null |
\caption{\label{EACC0sc}
Scalar products of five-vectors involved in the expression (\ref{Vect_eig}) for space components of eigenvector $\textbf{u}_\lambda$.}
| null |
\begin{ruledtabular}
\begin{tabular}{@{}l c c c c c@{}}
& $\mathbf{E}$ & $\mathbf{B}$ & $\mathbf{A}$ & $\mathbf{C}$ & $\mathbf{C}_0$ \\
\hline
$\mathbf{E}$ & $\mathbf{E}^2$ & ${\cal P}$ & $0$ & ${\cal Q}+\mathbf{B}^2$& $0$ \\
$\mathbf{B}$ & ${\cal P}$ & $\mathbf{B}^2$ & $0$ & ${\cal P}({\cal S}+\mathbf{E}^2)$ & $0$ \\
$\mathbf{A}$ & $0$ & $0$ &$\mathbf{E}^2({\cal S}+\mathbf{E}^2)-({\cal Q}+\mathbf{B}^2)$&$0$& $\mathbf{E}^2\mathbf{B}^2-{\cal P}^2$ \\
$\mathbf{C}$ & ${\cal Q}+\mathbf{B}^2$ & ${\cal P}({\cal S}+\mathbf{E}^2)$ & $0$ & ${\cal Q}\left({\cal S}+\mathbf{E}^2\right)+{\cal S}\mathbf{B}^2+{\cal P}^2$ & $0$ \\ %\hline
$\mathbf{C}_0$& $0$ & $0$ & $\mathbf{E}^2\mathbf{B}^2-{\cal P}^2$ & $0$ & $({\cal Q}+\mathbf{B}^2)\mathbf{B}^2-({\cal S}+\mathbf{E}^2){\cal P}^2$
\end{tabular}
\end{ruledtabular}
|
3
|
2008.03842
| null |
\caption[
|
\label{psr_table}
|
\begin{tabular}{rrrrrrrrrrr}
\toprule
& & & & & & & & & \\
{} & \multicolumn{1}{c}{$l$} & \multicolumn{1}{c}{$b$} & \multicolumn{1}{c}{$P$} & \multicolumn{1}{c}{$\dot{P}$} & \multicolumn{1}{c}{DM} & \multicolumn{1}{c}{$P_b$} & \multicolumn{1}{c}{$a$} & \multicolumn{1}{c}{$e$} & \multicolumn{1}{c}{$z$} & \multicolumn{1}{c}{Merger time} \\
PSR & (deg) & (deg) & (ms) & ($10^{-18}$ s/s) & (pc cm$^{-3}$) & (days) & (lt-s) & & (kpc) & (Gyr) \\
%& & & & & & & & & & \\
%\midrule
& & & & & & & & & \\
\midrule
& & & & Non-merging systems & & & & & \\
\midrule
& & & & & & & & & \\
J1518+4904 & 80.8 & 54.3 & 40.9 & 0.027 & 12 & 8.63 & 20.0 & 0.25 & 0.78 & 2400 \\
J0453+1559 & 184.1 & $-$17.1 & 45.8 & 0.19 & 30 & 4.07 & 14.5 & 0.11 & $-$0.15 & 1430 \\
J1811$-$1736 & 12.8 & 0.4 & 104.2 & 0.90 & 476 & 18.78 & 34.8 & 0.83 & 0.03 & 1000 \\
J1411+2551 & 33.4 & 72.1 & 62.4 & 0.096 & 12 & 2.62 & 9.2 & 0.17 & 1.08 & 460 \\
J1829+2456 & 53.3 & 15.6 & 41.0 & 0.052 & 14 & 1.18 & 7.2 & 0.14 & 0.24 & 60 \\
J1753$-$2240 & 6.3 & 1.7 & 95.1 & 0.97 & 159 & 13.64 & 18.1 & 0.30 & 0.09 & - \\
J1930$-$1852 & 20.0 & $-$16.9 & 185.5 & 18.0 & 43 & 45.06 & 86.9 & 0.40 & $-$0.58 & - \\
& & & & & & & & & & \\
\midrule
& & & & Merging systems & & & & & \\
\midrule
& & & & & & & & & \\
B1534+12 & 19.8 & 48.3 & 37.9 & 2.4 & 12 & 0.42 & 3.7 & 0.27 & 0.79 & 2.70 \\
J1756$-$2251 & 6.5 & 0.9 & 28.5 & 1.0 & 121 & 0.32 & 2.8 & 0.18 & 0.01 & 1.69 \\
J0509+3801 & 168.3 & $-$1.2 & 76.5 & 7.9 & 69 & 0.38 & 2.0 & 0.58 & $-$0.04 & 0.59 \\
J1913+1102 & 45.2 & 0.2 & 27.3 & 0.16 & 339 & 0.21 & 1.7 & 0.09 & 0.02 & 0.50 \\
J1906+0746 & 41.6 & 0.1 & 144.0 & 20000 & 218 & 0.17 & 1.4 & 0.08 & 0.02 & 0.30 \\
B1913+16 & 50.0 & 2.1 & 59.0 & 8.6 & 169 & 0.32 & 2.3 & 0.62 & 0.19 & 0.30 \\
J0737$-$3039A & 245.2 & $-$4.5 & 22.7 & 1.8 & 49 & 0.10 & 1.4 & 0.09 & $-$0.09 & 0.085 \\
J0737$-$3039B & 245.2 & $-$4.5 & 2773.5 & 890 & 49 & 0.10 & 1.5 & 0.09 & $-$0.09 & 0.085 \\
J1757$-$1854 & 10.0 & 2.9 & 21.5 & 2.6 & 378 & 0.18 & 2.2 & 0.60 & 0.37 & 0.076 \\
J1946+2052 & 57.7 & $-$2.0 & 16.9 & 0.90 & 94 & 0.08 & 1.1 & 0.06 & $-$0.14 & 0.046 \\
\bottomrule
\end{tabular}
|
4
|
2008.03842
| null |
\caption[
|
\label{result_table}
|
\begin{tabular}{ccccccc}
\toprule
PSR & $f_b$ & $\delta$ & $\tau_{\rm age}$ & $N_{\rm obs}$ & $N_{\rm pop}$ & $\mathcal{R}$ \\
& & & (Myr) & & & (Myr$^{-1}$) \\
\midrule
& & & & & & \\
B1534+12 & 6.0 & 0.04 & 208 & $98^{+455}_{-64}$ & $591^{+2750}_{-386}$ & $0.2^{+0.9}_{-0.1}$ \\
& & & & & & \\
J1756$-$2251 & 4.6 & 0.03 & 396 & $114^{+523}_{-80}$ & $523^{+2403}_{-367}$ & $0.3^{+1.4}_{-0.2}$ \\
& & & & & & \\
J1913+1102 & 4.6 & 0.06 & 2625 & $150^{+691}_{-104}$ & $688^{+3171}_{-477}$ & $0.2^{+1.0}_{-0.1}$ \\
& & & & & & \\
J0509+3801 & 4.6 & 0.06 & 710 & $186^{+838}_{-136}$ & $853^{+3849}_{-624}$ & $1.2^{+5.4}_{-0.9}$ \\
& & & & & & \\
J1906+0746 & 4.6 & 0.01 & 0.11 & $54^{+248}_{-32}$ & $248^{+1136}_{-147}$ & $4.1^{+19.1}_{-2.4}$ \\
& & & & & & \\
B1913+16 & 5.7 & 0.169 & 77 & $154^{+700}_{-108}$ & $880^{+4000}_{-617}$ & $2.4^{+10.8}_{-1.7}$ \\
& & & & & & \\
J0737$-$3039A & 2.0 & 0.27 & 159 & $342^{+1565}_{-252}$ & $683^{+3131}_{-503}$ & $2.9^{+13.0}_{-2.1}$ \\
& & & & & & \\
J1757$-$1854 & 4.6 & 0.06 & 87 & $162^{+739}_{-116}$ & $743^{+3391}_{-532}$ & $4.6^{+21}_{-3.3}$ \\
& & & & & & \\
J1946+2052 & 4.6 & 0.06 & 247 & $226^{+1034}_{-164}$ & $1036^{+4748}_{-751}$ & $3.5^{+16.2}_{-1.0}$ \\
& & & & & & \\
\bottomrule
\end{tabular}
|
5
|
2008.03842
| null |
\caption[
|
\label{param_range}
|
\begin{tabular}{cccc}
\toprule
Name of parameter & units & Minimum & Maximum \\
\midrule
Harmonic \m & -- & 1 & 5 \\
Survey integration time \tobs\ & seconds & 1 & $5 \times 10^3$\\
Mass of pulsar \ma\ & $M_{\odot}$ & 1 & 2.4 \\
Mass of companion \mb\ & $M_{\odot}$ & 0.2 & $10^9$ \\
Spin period of pulsar \ps\ & seconds & $10^{-3}$ & 5 \\
Inclination of binary system \inc\ & degrees & $0^{\circ}$ & $90^{\circ}$ \\
Angle of periastron passage \om\ & degrees & $0^{\circ}$ & $360^{\circ}$ \\
Eccentricity \ecc & -- & 0 & 0.9 \\
Orbital period \pb\ & days & $10^{-3}$ & $10^{3}$ \\
\bottomrule
\end{tabular}
|
6
|
2008.03842
| null |
\caption[
|
\label{survey_table}
|
\begin{tabular}{ccccccc}
\toprule
Survey & Gain $G$ & Center Frequency $f_{\rm c}$ & Bandwidth $B$ & System temperature $T_{\rm sys}$ & Integration time $t_{\rm int}$ \\
-- & (K/Jy) & (MHz) & (MHz) & (K) & (s) \\
\midrule
PALFA\footnote{Pulsar Arecibo L-band Feed Array, \citet{PALFA}} & 8.5 & 1374 & 300 & 25 & 268 \\
PMSURV\footnote{Parkes Multibeam SURvey, \citet{PMSURV}} & 0.6 & 1374 & 288 & 25 & 2100 \\
AODRIFT\footnote{ArecibO DRIFT scan survey, \citet{aodrift_1}} & 10 & 327 & 25 & 100 & 50 \\
GBNCC\footnote{Green Bank North Celestial Cap Survey, \citet{gbncc}} & 2 & 350 & 100 & 46 & 120 \\
HTRU--low\footnote{High Time Resolution Universe low-latitude survey \citet{htru_low_mid}} & 0.6 & 1352 & 340 & 25 & 340 \\
HTRU--mid\footnote{High Time Resolution Universe mid-latitude survey \citet{htru_low_mid}} & 0.6 & 1352 & 340 & 25 & 540 \\
\bottomrule
\end{tabular}
|
7
|
2008.03846
| null |
\caption{Representatives of equivalence classes in ${\mathcal B}_{\alpha}$ \eqref{eq:Ca}}
|
\label{tab:a}
|
\begin{tabular}{ccccccc}
\hline
\cellcolor[HTML]{FFCE93}& \cellcolor[HTML]{FFCE93}$\alpha_{11}$ &\cellcolor[HTML]{FFCE93} $\alpha_{21}$ &\cellcolor[HTML]{FFCE93} $\alpha_{31}$ & \cellcolor[HTML]{FFCE93}$\alpha_{12}$ &\cellcolor[HTML]{FFCE93} $\alpha_{22}$ &\cellcolor[HTML]{FFCE93} $\alpha_{32}$ \\ \hline
(\newrowc\label{row:c1}) & $2$ & $1$ & $1$ & $1$ & $0$ & $0$ \\ \hline
(\newrowc\label{row:c2.1}) & $2$ & $2$ & $0$ & $1$ & $0$ & $1$ \\ \hline
(\newrowc\label{row:c2.2}) & $1$ & $3$ & $0$ & $0$ & $1$ & $1$ \\ \hline
(\newrowc\label{row:c2.3}) & $1$ & $2$ & $1$ & $0$ & $0$ & $2$ \\ \hline
(\newrowc\label{row:c3.1}) & $2$ & $2$ & $0$ & $1$ & $0$ & $2$ \\ \hline
(\newrowc\label{row:c3.2}) & $1$ & $3$ & $0$ & $0$ & $1$ & $2$ \\ \hline
(\newrowc\label{row:c3.3}) & $1$ & $2$ & $1$ & $0$ & $0$ & $3$ \\ \hline
(\newrowc\label{row:c4.1}) & $2$ & $2$ & $0$ & $1$ & $0$ & $3$ \\ \hline
(\newrowc\label{row:c4.2}) & $1$ & $3$ & $0$ & $0$ & $1$ & $3$ \\ \hline
(\newrowc\label{row:c4.3}) & $1$ & $2$ & $1$ & $0$ & $0$ & $4$ \\ \hline
(\newrowc\label{row:c5.1}) & $4$ & $0$ & $0$ & $1$ & $1$ & $1$ \\ \hline
(\newrowc\label{row:c5.2}) & $3$ & $1$ & $0$ & $0$ & $2$ & $1$ \\ \hline
\end{tabular}
|
16
|
2508.00329
| null |
\caption{Various $h$ and corresponding Degrees of Freedoms (DoF) in the mesh independence test.}
|
\label{tab2}
|
\begin{tabular}{ccccccc}
\hline
\hline
$m$ & 0.05 & 0.1 & 0.2 & 0.4 & 0.8 & 1.6 \\
DoF &3272056&3270616&3243616&3220864&3204808&3201952\\
\hline
\end{tabular}
|
114
|
2508.00575
| null | null |
\label{thm:consequences:future-answering}
| null |
8
|
2008.03846
| null |
\caption{All candidates for multistable networks in ${\mathcal G}_0$ with $4$ reactants and $3$ species}
|
\label{tab:net}
|
\begin{tabular}{cc|r}
\hline
\cellcolor[HTML]{FFCE93} &\cellcolor[HTML]{FFCE93} Network
&\cellcolor[HTML]{FFCE93} $ (\beta_{11}, \beta_{21}, \beta_{31}, \beta_{12}, \beta_{22}, \beta_{32})\in {\mathbb Z}^6_{\geq 0} $
\\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r1}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ 2X_1 + X_2 + X_3 \rightarrow 3X_1 $ \\
$ X_1 \rightarrow X_2 + X_3 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r2.1}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ 2X_1 + 2X_2 \rightarrow \beta_{11}X_1 + \beta_{21}X_2 + \beta_{31}X_3 $ \\
$ X_1 + X_3 \rightarrow \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$ (\beta_{11}, -\beta_{22}(\beta_{11}-2)+2, \beta_{11}-2, 0, \beta_{22}, 0) $ \\
$ \beta_{11} \in\{3,4\}; \;\;\;\;0<\beta_{22}{\le}\frac{2}{\beta_{11}-2} $
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
\cellcolor[HTML]{6CDEFF}(\newrowa\label{row:2r2.2}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 3X_2 \rightarrow 4X_2 + X_3$ \\
$ X_2 + X_3 \rightarrow X_1 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r2.3}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 2X_2 + X_3 \rightarrow \beta_{21}X_2 $ \\
$ 2X_3 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 + \beta_{32}X_3 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$ (0, \beta_{21}, 0, \beta_{12}, \beta_{12}(2-\beta_{21}), \beta_{12}+2 ) $ \\
$ \beta_{21} \in \{ 0,1\}; \;\;\;\;\beta_{12}>0 $
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r3.1}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ 2X_1 + 2X_2 \rightarrow \beta_{11}X_1 + \beta_{21}X_2 + \beta_{31}X_3 $ \\
$ X_1 + 2X_3 \rightarrow \beta_{22}X_2 + \beta_{32}X_3 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$ (\beta_{11}, -\beta_{22}(\beta_{11}-2)+2, (2-\beta_{32})(\beta_{11}-2), 0, \beta_{22}, \beta_{32} )$ \\
$ \beta_{11} \in \{ 3,4\};\;\;\;\;0<\beta_{22}{\le}\frac{2}{\beta_{11}-2};\;\;\;\; \beta_{32}\in \{0, 1\} $
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r3.2}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 3X_2 \rightarrow 4X_2 + \beta_{31}X_3 $ \\
$ X_2 + 2X_3 \rightarrow X_1 + \beta_{32}X_3 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$ (0, 4, \beta_{31}, 1, 0, 2-\beta_{31} )$ \\
$ \beta_{31} \in \{1,2\} $
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
\cellcolor[HTML]{6CDEFF}(\newrowa\label{row:2r3.3}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 2X_2 + X_3 \rightarrow \beta_{21}X_2 $ \\
$ 3X_3 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 + \beta_{32}X_3 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$(0, \beta_{21}, 0, \beta_{12}, \beta_{12}(2-\beta_{21}), \beta_{12}+3 )$ \\
$ \beta_{21} \in \{0,1\};\;\;\;\; \beta_{12}>0 $
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r4.1}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ 2X_1 + 2X_2 \rightarrow \beta_{11}X_1 + \beta_{21}X_2 + \beta_{31}X_3 $ \\
$ X_1 + 3X_3 \rightarrow \beta_{22}X_2 + \beta_{32}X_3 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$ (\beta_{11}, -\beta_{22}(\beta_{11}-2)+2, (3-\beta_{32})(\beta_{11}-2), 0, \beta_{22}, \beta_{32} )$ \\
$ \beta_{11} \in \{3,4\};\;\;\;\;0<\beta_{22}{\le}\frac{2}{\beta_{11}-2};\;\;\;\; \beta_{32}\in \{0, 1, 2\} $
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r4.2}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 3X_2 \rightarrow 4X_2 + \beta_{31}X_3 $ \\
$ X_2 + 3X_3 \rightarrow X_1 + \beta_{32}X_3 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$ (0, 4, \beta_{31}, 1, 0, 3-\beta_{31} )$ \\
$ \beta_{31} \in \{1,2,3\} $
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r4.3}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 2X_2 + X_3 \rightarrow \beta_{21}X_2 $ \\
$ 4X_3 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 + \beta_{32}X_3 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$(0, \beta_{21}, 0, \beta_{12}, \beta_{12}(2-\beta_{21}), \beta_{12}+4 )$ \\
$\beta_{21} \in \{0,1\};\;\;\;\; \beta_{12}>0 $
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r5.1}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ 4X_1 \rightarrow \beta_{11}X_1 + \beta_{21}X_2 + \beta_{31}X_3 $ \\
$ X_1 + X_2 + X_3 \rightarrow 0 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$ (\beta_{21}+4, \beta_{21}, \beta_{21}, 0, 0, 0 )$ \\
$\beta_{21}>0$
\end{tabular} \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowa\label{row:2r5.2}) \\ \quad
\end{tabular} &
\begin{tabular}[c]{@{}c@{}}
$ 3X_1 + X_2 \rightarrow \beta_{11}X_1 + X_3 $ \\
$ 2X_2 + X_3 \rightarrow \beta_{12}X_1 + 3X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$(\beta_{11}, 0, 1, 3-\beta_{11}, 3, 0 ) $ \\
$ \beta_{11} \in \{0,1,2\} $
\end{tabular} \\ \hline
\end{tabular}
|
9
|
2008.03846
| null |
\caption{Representatives of equivalence classes in ${\mathcal C}_{\sigma}$ \eqref{eq:Csigma}}
|
\label{tab:sigma}
|
\begin{tabular}{c|c|c|c}
\hline
% & $\left(\alpha_{11},\alpha_{21},\beta_{11},\beta_{21},\alpha_{12},\alpha_{22}\right)$ & Network & $\beta_{i2}$ possible value \\ \hline
\cellcolor[HTML]{FFCE93}
\begin{tabular}[r]{@{}r@{}}
% $\alpha_{11}+\alpha_{21}=3 $ \\
% $\beta_{11}+\beta_{21}<3 $ \\
% $\alpha_{21}+\alpha_{22}=3 $\\
see ``reverse1.mw"
\end{tabular}
&
\cellcolor[HTML]{FFCE93}
\begin{tabular}[r]{@{}r@{}}
% $\alpha_{11}+\alpha_{21}<3 $ \\
% $\beta_{11}+\beta_{21}<3 $ \\
% $\alpha_{21}+\alpha_{22}=3 $\\
see ``reverse2.mw"
\end{tabular}
&
\cellcolor[HTML]{FFCE93}
\begin{tabular}[r]{@{}r@{}}
%$\alpha_{11}+\alpha_{21}=3 $ \\
%$\beta_{11}+\beta_{21}<3 $ \\
%$\alpha_{21}+\alpha_{22}<3 $\\
see ``reverse3.mw"
\end{tabular}
&
\cellcolor[HTML]{FFCE93}
\begin{tabular}[r]{@{}r@{}}
%$\alpha_{11}+\alpha_{21}=3 $ \\
%$\beta_{11}+\beta_{21}=3 $ \\
%$\alpha_{21}+\alpha_{22}\leq 3 $\\
see ``reverse4.mw"
\end{tabular}
\\ \hline
$\left(
\begin{array}{ccc}
0&1&3\\
3&0&0
\end{array}
\right)$&
\cellcolor[HTML]{FFE7FD}$\left(
\begin{array}{ccc}
{\mathbf
0}&{\mathbf
1}&{\mathbf
3}\\
{\mathbf
0}&{\mathbf
1}&{\mathbf
0}
\end{array}
\right)$
&
$\left(
\begin{array}{ccc}
0&1&2\\
3&0&0
\end{array}
\right)$
&
$\left(
\begin{array}{ccc}
0&1&2\\
3&2&0
\end{array}
\right)$
\\ \hline
$\left(
\begin{array}{ccc}
0&1&2\\
3&0&1
\end{array}
\right)$&
$ \cellcolor[HTML]{FFE7FD}\left(
\begin{array}{ccc}
{\mathbf
0}&{\mathbf
1}&{\mathbf
2}\\
{\mathbf
0}&{\mathbf
1}&{\mathbf
1}
\end{array}
\right)$
&
$\left(
\begin{array}{ccc}
0&1&2\\
3&1&0
\end{array}
\right)$
&
$\left(
\begin{array}{ccc}
0&1&3\\
3&2&0
\end{array}
\right)$
\\\hline
$\left(
\begin{array}{ccc}
0&1&3\\
3&1&0
\end{array}
\right)$ &
$\left(
\begin{array}{ccc}
0&1&3\\
1&0&0
\end{array}
\right)$
&
\cellcolor[HTML]{FFE7FD}
$\left(
\begin{array}{ccc}
{\mathbf
1}&{\mathbf
0}&{\mathbf
2}\\
{\mathbf
2}&{\mathbf
0}&{\mathbf
0}
\end{array}
\right)$
&
$\left(
\begin{array}{ccc}
0&2&3\\
3&1&0
\end{array}
\right)$
\\ \hline
$\left(
\begin{array}{ccc}
0&2&3\\
2&0&0
\end{array}
\right)$&
\cellcolor[HTML]{FFE7FD}
$\left(
\begin{array}{ccc}
{\mathbf
0}&{\mathbf
1}&{\mathbf
2}\\
{\mathbf
1}&{\mathbf
0}&{\mathbf
1}
\end{array}
\right)$
&
\cellcolor[HTML]{FFE7FD}$\left(
\begin{array}{ccc}
{\mathbf
1}&{\mathbf
0}&{\mathbf
2}\\
{\mathbf
2}&{\mathbf
1}&{\mathbf
0}
\end{array}
\right)$
&
\\\hline
$ \cellcolor[HTML]{FFE7FD}\left(
\begin{array}{ccc}
{\mathbf
2}&{\mathbf
0}&{\mathbf
3}\\
{\mathbf
1}&{\mathbf
0}&{\mathbf
0}
\end{array}
\right)$&
$\left(
\begin{array}{ccc}
0&2&3\\
1&0&0
\end{array}
\right)$
&&
\\ \hline
$ \cellcolor[HTML]{FFE7FD}\left(
\begin{array}{ccc}
{\mathbf
1}&{\mathbf
0}&{\mathbf
3}\\
{\mathbf
2}&{\mathbf
0}&{\mathbf
0}
\end{array}
\right)$&
$\left(
\begin{array}{ccc}
1&0&3\\
0&2&0
\end{array}
\right)$&&
\\ \hline
$ \cellcolor[HTML]{FFE7FD}\left(
\begin{array}{ccc}
{\mathbf
1}&{\mathbf
0}&{\mathbf
2}\\
{\mathbf
2}&{\mathbf
0}&{\mathbf
1}
\end{array}
\right)$&
$ \cellcolor[HTML]{FFE7FD}\left(
\begin{array}{ccc}
{\mathbf
1}&{\mathbf
0}&{\mathbf
2}\\
{\mathbf
0}&{\mathbf
2}&{\mathbf
1}
\end{array}
\right)$
&&
\\ \hline
$ \cellcolor[HTML]{FFE7FD}\left(
\begin{array}{ccc}
{\mathbf
1}&{\mathbf
0}&{\mathbf
3}\\
{\mathbf
2}&{\mathbf
1}&{\mathbf
0}
\end{array}
\right)$&
$\left(
\begin{array}{ccc}
1&0&3\\
1&2&0
\end{array}
\right)$
&&
\\ \hline
$\left(
\begin{array}{ccc}
2&0&3\\
1&2&0
\end{array}
\right)$&
$\left(
\begin{array}{ccc}
2&0&3\\
0&2&0
\end{array}
\right)$
&&
\\ \hline
\end{tabular}
|
10
|
2008.03846
| null |
\caption{All candidates for multistable networks in ${\mathcal G}_1$ with $3$ reactants and $2$ species}
|
\label{tab:net2}
|
\begin{tabular}{cc|c|r}
\hline
\cellcolor[HTML]{FFCE93}&\cellcolor[HTML]{FFCE93} $\left(\alpha_{11},\alpha_{21},\beta_{11},\beta_{21},\alpha_{12},\alpha_{22}\right)$ & \cellcolor[HTML]{FFCE93}Network & \cellcolor[HTML]{FFCE93} $\beta_{12}$ and $\beta_{22}$ in ${\mathbb Z}_{\geq 0}$ \\ \hline
%\begin{tabular}[c]{@{}c@{}}
% (\newrowb\label{row:1r1i3}) \\ \quad
% \end{tabular} & $(0,3,1,0,2,1)$ &
% \begin{tabular}[c]{@{}c@{}}
% $ 3X_2 \Leftrightarrow X_1 $ \\
% $ 2X_1 + X_2 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
% \end{tabular} &
%\begin{tabular}[r]{@{}r@{}}
% No integer value for $\beta_{21}$, \\
% see the proof
%$\beta_{22}=-3(\beta_{12}-2)+1 $ \\
%$\beta_{12}-2>0$
% \end{tabular}
% \\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i4}) \\ \quad
\end{tabular} & $(2,1,0,0,3,0)$ &
\begin{tabular}[c]{@{}c@{}}
$ 2X_1 + X_2 \Leftrightarrow 0 $ \\
$ 3X_1 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=\frac{1}{2}(\beta_{12}-3) $ \\
$\beta_{12}-3>0$
\end{tabular}
\\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i5}) \\ \quad
\end{tabular} & $(1,2,0,0,3,0)$ &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 2X_2 \Leftrightarrow 0 $ \\
$ 3X_1 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=2(\beta_{12}-3) $ \\
$\beta_{12}-3>0$
\end{tabular}
\\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i6}) \\ \quad
\end{tabular} & $(1,2,0,0,2,1)$ &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 2X_2 \Leftrightarrow 0 $ \\
$ 2X_1 + X_2 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=2(\beta_{12}-2)+1 $ \\
$\beta_{12}-2>0$
\end{tabular}
\\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i7}) \\ \quad
\end{tabular} & $(1,2,0,1,3,0)$ &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 2X_2 \Leftrightarrow X_2 $ \\
$ 3X_1 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=\beta_{12}-3$ \\
$\beta_{12}-3>0$
\end{tabular}
\\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i8}) \\ \quad
\end{tabular} & $(0,0,1,1,3,0)$ &
\begin{tabular}[c]{@{}c@{}}
$ 0 \Leftrightarrow X_1 + X_2 $ \\
$ 3X_1 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=\beta_{12}-3 $ \\
$\beta_{12}-3>0$
\end{tabular}
\\ \hline
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i9}) \\ \quad
\end{tabular} & $(0,0,1,1,2,1)$ &
\begin{tabular}[c]{@{}c@{}}
$ 0 \Leftrightarrow X_1 + X_2 $ \\
$ 2X_1 + X_2 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=(\beta_{12}-2)+1 $ \\
$\beta_{12}-2>0$
\end{tabular}
\\ \hline
\cellcolor[HTML]{6CDEFF}
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i10}) \\ \quad
\end{tabular} & $(0,1,1,0,2,1)$ &
\begin{tabular}[c]{@{}c@{}}
$ X_2 \Leftrightarrow X_1 $ \\
$ 2X_1 + X_2 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=-(\beta_{12}-2)+1 $ \\
$\beta_{12}-2>0$
\end{tabular}
\\ \hline
\cellcolor[HTML]{6CDEFF}
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i11}) \\ \quad
\end{tabular} & $(1,0,0,2,2,1)$ &
\begin{tabular}[c]{@{}c@{}}
$ X_1 \Leftrightarrow 2X_2 $ \\
$ 2X_1 + X_2 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=-2(\beta_{12}-2)+1 $ \\
$\beta_{12}-2>0$
\end{tabular}
\\ \hline
\cellcolor[HTML]{6CDEFF}
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i1}) \\ \quad
\end{tabular} & $(1,2,0,0,2,0)$&
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 2X_2 \Leftrightarrow 0 $ \\
$ 2X_1 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular}&
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=2(\beta_{12}-2) $ \\
$\beta_{12}-2>0$
\end{tabular}
\\ \hline
\cellcolor[HTML]{6CDEFF}
\begin{tabular}[c]{@{}c@{}}
(\newrowb\label{row:1r1i2}) \\ \quad
\end{tabular} & $(1,2,0,1,2,0)$ &
\begin{tabular}[c]{@{}c@{}}
$ X_1 + 2X_2 \Leftrightarrow X_2 $ \\
$ 2X_1 \rightarrow \beta_{12}X_1 + \beta_{22}X_2 $
\end{tabular} &
\begin{tabular}[r]{@{}r@{}}
$\beta_{22}=\beta_{12}-2 $ \\
$\beta_{12}-2>0$
\end{tabular}
\\ \hline
\end{tabular}
|
11
|
2008.03846
| null |
\caption{Supporting Information Files}
|
\label{tab:sup}
|
\begin{tabular}{||l c l||}
\hline
\cellcolor[HTML]{FFCE93} Name &\cellcolor[HTML]{FFCE93} File Type & \cellcolor[HTML]{FFCE93}Results \\ [0.5ex]
\hline\hline
1. \texttt{WitnessForMultistableNetworksInTheorem2.6.mw/.pdf} & \texttt{Maple/PDF} & Theorem \ref{thm:g1}\\
2. \texttt{EquivalenceClassesOfSmallReversibleNetworks4.mw/.pdf} & \texttt{Maple/PDF} & Lemma \ref{lm:g1} \\
3. \texttt{EquivalenceClassesOfSmallReversibleNetworks3.mw/.pdf} & \texttt{Maple/PDF}& Lemma \ref{lm:g1} \\
4. \texttt{EquivalenceClassesOfSmallReversibleNetworks2.mw/.pdf} & \texttt{Maple/PDF} & Lemma \ref{lm:g1} \\
5. \texttt{EquivalenceClassesOfSmallReversibleNetworks1.mw/.pdf} & \texttt{Maple/PDF} & Lemma \ref{lm:g1} \\
%\texttt{reverse5.mw} & \texttt{Maple} & Lemma \ref{lm:g1} \\
%\texttt{rrc.mw} & \texttt{Maple} & Theorem \ref{thm:g1}\\
6. \texttt{WitnessForMultistableNetworksInTheorem2.4.mw/.pdf} & \texttt{Maple/PDF} & Theorem \ref{thm:4-reactant}\\
7. \texttt{EquivalenceClassesOfSmallIrreversibleNetworks.mw/.pdf} & \texttt{Maple/PDF} & Lemma \ref{lm:4-reactant}\\
\hline
\end{tabular}
|
12
|
2508.00602
| null |
\caption{Static LeakSealer approach extrinsic evaluation.}
|
\label{tab:static-results}
|
\begin{tabular}{l|c|c|c}
\hline
\textbf{Metric} & \textbf{OpenAI Dataset} & \textbf{ToxicChat Dataset} & \textbf{PII Dataset} \\
\hline
Purity & 0.85 & 0.97 & 0.77 \\
Accuracy & 0.84 & 0.96 & 0.75 \\
Precision & 0.73 & 0.79 & 0.68 \\
Recall & 0.75 & 0.75 & 0.88 \\
F1-score & 0.74 & 0.77 & 0.77 \\
\hline
\end{tabular}
|
13
|
2508.00602
| null |
\caption{Precision, Recall, and F1 Score of Static LeakSealer against baselines.}
|
\label{table:static-pr-results}
|
\begin{tabular}{c|ccc|ccc|ccc}
\hline
\textbf{Model} & \multicolumn{3}{c|}{\textbf{OpenAI}} & \multicolumn{3}{c|}{\textbf{ToxicChat}} & \multicolumn{3}{c}{\textbf{PII}} \\
\hline
& \textbf{Precision} & \textbf{Recall} & \textbf{F1} & \textbf{Precision} & \textbf{Recall} & \textbf{F1} & \textbf{Precision} & \textbf{Recall} & \textbf{F1} \\
\hline
\textbf{Llama Guard 3} & 0.79 & 0.78 & {\textcolor{highlightnumber}{\textbf{0.79}}} & 0.74 & 0.38 & 0.50 & 0.82 & 0.46 & 0.59 \\
\textbf{Judge (DeepSeek-R1)} & 0.55 & 0.68 & 0.61 & 0.65 & 0.58 & 0.61 & 0.58 & 0.48 & 0.53 \\
\textbf{Judge (GPT-4o)} & 0.64 & 0.95 & 0.77 & 0.74 & 0.49 & 0.59 & 0.81 & 0.64 & 0.72 \\
\textbf{Judge (Ministral)} & 0.64 & 0.93 & 0.76 & 0.67 & 0.56 & 0.61 & 0.62 & 0.65 & 0.64 \\
\textbf{Judge (Llama 3.1)} & 0.54 & 0.93 & 0.69 & 0.60 & 0.41 & 0.49 & 0.70 & 0.55 & 0.62 \\
\hline
\hline
\textbf{LeakSealer} & 0.73 & 0.75 & 0.74 & 0.79 & 0.75 & {\textcolor{highlightnumber}{\textbf{0.77}}} & 0.68 & 0.88 & {\textcolor{highlightnumber}{\textbf{0.77}}} \\
\hline
\end{tabular}
|
14
|
2508.00602
| null |
\caption{Evaluation results in the Dynamic setting. AUPRC does not apply to LLM-As-A-Judge techniques.}
|
\label{table:combined_results_dynamic}
|
\begin{tabular}{p{2cm}|l|l|l|l|l|l}
\hline
\textbf{Dataset} & \textbf{Technique} & \textbf{Acc.} & \textbf{Prec.} & \textbf{Rec.} & \textbf{F1} & \textbf{AUPRC} \\
\hline
\multirow{6}{*}{OpenAI} & Llama Guard 3 & \textbf{0.86} & \textbf{0.77} & 0.79 & 0.78 & \textbf{0.90} \\
& Judge (DeepSeek-R1) & 0.74 & 0.59 & 0.65 & 0.62 & -- \\
& Judge (GPT-4o) & 0.83 & 0.66 & \textbf{0.94} & 0.78 & -- \\
& Judge (Llama 3.1) & 0.72 & 0.54 & 0.92 & 0.68 & -- \\
& Judge (Ministral) & 0.84 & 0.70 & 0.91 & \textbf{0.79} & -- \\
& LeakSealer & 0.83 & 0.73 & 0.75 & 0.74 & 0.83 \\
\hline
\multirow{6}{*}{ToxicChat} & Llama Guard 3 & 0.94 & 0.66 & 0.38 & 0.48 & 0.56 \\
& Judge (DeepSeek-R1) & 0.95 & 0.63 & 0.59 & 0.61 & -- \\
& Judge (GPT-4o) & 0.95 & 0.69 & 0.46 & 0.55 & -- \\
& Judge (Llama 3.1) & 0.93 & 0.50 & 0.35 & 0.41 & -- \\
& Judge (Ministral) & 0.95 & 0.64 & 0.58 & 0.61 & -- \\
& LeakSealer & \textbf{0.96} & \textbf{0.84} & \textbf{0.63} & \textbf{0.72} & \textbf{0.76} \\
\hline
\multirow{6}{*}{PII} & Llama Guard 3 & 0.64 & \textbf{0.91} & 0.31 & 0.46 & 0.84 \\
& Judge (DeepSeek-R1) & 0.58 & 0.60 & 0.46 & 0.52 & -- \\
& Judge (GPT-4o) & 0.75 & 0.87 & 0.59 & 0.70 & -- \\
& Judge (Llama 3.1) & 0.66 & 0.76 & 0.45 & 0.57 & -- \\
& Judge (Ministral) & 0.60 & 0.60 & 0.59 & 0.59 & -- \\
& LeakSealer & \textbf{0.91} & 0.88 & \textbf{0.95} & \textbf{0.92} & \textbf{0.97} \\
\hline
\end{tabular}
|
15
|
2508.00329
| null |
\caption{Parameter values or ranges selected for this study.}
|
\label{tab1}
|
\begin{tabular}{cc}
\hline
\textbf{Parameters} & \textbf{Values} \\
\hline
$m$ & 0.1 \\
Number of Degrees of Freedom & 3270616\\
$r_0$ & 0.5 \\
$R$ & 700 \\
$Q$ & [10, 200] \\
$\delta$ & [1, 10] \\
\hline
\end{tabular}
|
115
|
2508.00575
| null | null |
\label{part:linear:combined}
| null |
8,826
|
2501.14370
| null |
\caption{The Summary of Kernel Performance and Energy Efficiency.}
|
\label{tab_kernel}
| null |
17
|
2508.05628v1
| null |
\caption{Full 1.58B-token training corpus with ZWNJ statistics.}
|
\label{tab:data_full}
|
\begin{tabular}{lrrr}
\toprule
Dataset & Tokens & Genre & ZWNJ\% \\
\midrule
Hamshahri-2 & 190M & News & 12.3 \\
Wikipedia-fa & 415M & Encyclopedia & 11.2 \\
VOA Persian & 285M & News & 9.8 \\
MirasText & 310M & Literature & 14.1 \\
Ganjoor & 178M & Poetry & 16.8 \\
Twitter-fa & 89M & Social Media & 3.2 \\
Scientific-fa & 111M & Academic & 15.6 \\
\bottomrule
\end{tabular}
|
18
|
2508.05628v1
| null |
\caption{Main results averaged over 3 runs. H-Net++ achieves state-of-the-art across all metrics. Seg F1 is only applicable to models with learnable or byte-level segmentation.}
|
\label{tab:main_detailed}
|
\begin{tabular}{lcccc}
\toprule
Model & BPB$\downarrow$ & ParsGLUE$\uparrow$ &
Robustness$\uparrow$ & Seg F1$\uparrow$ \\ \midrule
GPT‑2‑fa & 1.342 & 71.2 & 45.3 & – \\
ParsBERT & – & 73.8 & 52.1 & – \\
mT5-small & 1.387 & 70.5 & 49.8 & – \\
ByT5‑fa & 1.425 & 68.9 & 61.2 & 52.3 \\
MegaByte‑fa& 1.398 & 69.4 & 58.7 & 31.2 \\
H-Net-Base & 1.256 & 74.1 & 64.2 & 68.4 \\
\midrule
H‑Net++ & \textbf{1.183} & \textbf{76.6} &
\textbf{69.4} & \textbf{73.8} \\
\bottomrule
\end{tabular}
|
19
|
2508.05628v1
| null |
\caption{ParsGLUE breakdown showing consistent improvements.}
|
\label{tab:parsglue_breakdown}
|
\begin{tabular}{lccccc}
\toprule
Model & Sentiment & NLI & NER & QA & Avg \\ \midrule
GPT‑2‑fa & 68.3 & 72.1 & 74.5 & 69.8 & 71.2 \\
ParsBERT & 71.2 & 75.3 & 76.1 & 72.4 & 73.8 \\
H‑Net++ & \textbf{74.8} & \textbf{78.2} & \textbf{79.3} & \textbf{74.1} & \textbf{76.6} \\
\bottomrule
\end{tabular}
|
20
|
2508.05628v1
| null |
\caption{Ablation study. The Transformer mixer provides the largest gain.}
|
\label{tab:ablation_detailed}
|
\begin{tabular}{lcc}
\toprule
\textbf{Configuration} & BPB$\uparrow$ & ParsGLUE$\downarrow$ \\ \midrule
Full H‑Net++ & 1.183 & 76.6 \\
\quad – Transformer Mixer & 1.256 (+0.073) & 75.4 (–1.2) \\
\quad – Hyper‑prior& 1.224 (+0.041) & 75.8 (–0.8) \\
\quad – ZWNJ Embedding & 1.208 (+0.025) & 75.9 (–0.7) \\
\quad – Morphology Loss & 1.201 (+0.018) & 76.1 (–0.5) \\
\quad – Curriculum & 1.195 (+0.012) & 76.2 (–0.4) \\
\bottomrule
\end{tabular}
|
21
|
2508.05628v1
| null |
\caption{Morphological segmentation accuracy.}
|
\label{tab:seg_eval}
|
\begin{tabular}{lccc}
\toprule
Model & Precision & Recall & F1 \\ \midrule
ByT5-fa & 48.7 & 56.2 & 52.3 \\
MegaByte-fa & 35.4 & 27.8 & 31.2 \\
H-Net-Base & 71.2 & 65.8 & 68.4 \\
H-Net++ & \textbf{76.3} & \textbf{71.5} & \textbf{73.8} \\
\bottomrule
\end{tabular}
|
22
|
2508.05628v1
| null |
\caption{Distribution of learned chunk types in H-Net++.}
|
\label{tab:chunk_stats}
|
\begin{tabular}{lcc}
\toprule
Chunk Type & Avg Length (bytes) & Frequency \\ \midrule
Simple words & 5.2 & 42\% \\
Compound words & 11.3 & 28\% \\
With clitics & 8.7 & 19\% \\
Punctuation & 1.8 & 11\% \\
\bottomrule
\end{tabular}
|
23
|
2508.05628v1
| null |
\caption{Runtime efficiency metrics on A100 GPU.}
|
\label{tab:efficiency}
|
\begin{tabular}{lccc}
\toprule
Model & Memory (GB) & FLOPs/token & Latency (ms) \\ \midrule
GPT-2-fa & 38.2 & 124M & 12.3 \\
ByT5-fa & 71.3 & 892M & 45.7 \\
H-Net++ & 43.1 & 198M & 18.4 \\
\bottomrule
\end{tabular}
|
24
|
2508.05628v1
| null |
\caption{H-Net++ scaling results. L=layers, H=hidden size. Larger models show diminishing returns beyond 500M parameters.}
| null |
\begin{tabular}{@{}lccccc@{}}
\toprule
Model & Params & L & H & BPB & ParsGLUE \\
\midrule
Small & 125M & 3 & 768 & 1.241 & 74.2 \\
Base & 252M & 3 & 1024 & 1.183 & 76.6 \\
Large & 500M & 4 & 1280 & 1.152 & 77.9 \\
XL & 1.1B & 4 & 1600 & 1.121 & 79.3 \\
\bottomrule
\end{tabular}
|
25
|
2508.05628v1
| null |
\caption{Zero-shot cross-lingual performance. The model transfers best to closely related languages (Dari) and struggles with different morphological systems (Turkish).}
| null |
\begin{tabular}{lccc}
\toprule
Target Language & Segmentation F1 & BPB & Degradation \\
\midrule
Persian (in-domain) & 73.8 & 1.183 & -- \\
Dari & 68.2 & 1.287 & -7.6\% \\
Tajik (Cyrillic) & 45.3 & 1.893 & -38.6\% \\
Urdu & 52.1 & 1.562 & -29.4\% \\
Arabic & 41.7 & 1.734 & -43.5\% \\
Turkish & 38.9 & 1.821 & -47.3\% \\
\bottomrule
\end{tabular}
|
26
|
2508.05628v1
| null |
\caption{Performance breakdown by morphological phenomenon, showing strong performance on productive affixes.}
| null |
\begin{tabular}{lcc}
\toprule
Phenomenon & Precision & Recall \\
\midrule
Compound words & 82.3 & 79.1 \\
Plural markers (-hā) & 91.2 & 88.7 \\
Ezafe construction & 78.4 & 71.3 \\
Verbal prefixes (mi-, be-) & 88.9 & 92.1 \\
Object markers (-rā) & 85.6 & 81.2 \\
Comparative (-tar) & 79.3 & 76.8 \\
Clitics & 74.2 & 69.5 \\
\bottomrule
\end{tabular}
|
27
|
2508.05628v1
| null |
\caption{H-Net++ segmentation examples. N=noun, PL=plural, EZ=ezafe, PRO=pronoun, ASP=aspect, V=verb, COMP=compound, NEG=negation, ADJ=adjective, SUP=superlative, DER=derivational.}
| null |
\begin{tabular}{@{}lll@{}}
\toprule
Input & H-Net++ Chunks & Type \\
\midrule
ketab-hā-ye man & [ketab] [hā] [ye] [man] & N+PL+EZ+PRO \\
mi-nevis-am & [mi] [nevis] [am] & ASP+V+1SG \\
dānesh-gāh-e & [dānesh] [gāh] [e] & COMP+EZ+\\
\quad tehrān & [tehrān] & N \\
na-mi-tavān-ad & [na] [mi] [tavān] [ad] & NEG+ASP+V+3SG \\
bozorg-tar-in & [bozorg] [tar] [in] & ADJ+COMP+SUP \\
ketāb-forush-i-hā & [ketāb] [forush] [i] [hā] & COMP+DER+PL \\
\bottomrule
\end{tabular}
|
28
|
2508.05628v1
| null |
\caption{Common error patterns showing challenges with Arabic loanwords and non-textual content.}
| null |
\begin{tabular}{@{}lll@{}}
\toprule
Input & H-Net++ Output & Error Type \\
\midrule
āzmāyeshgāh & [āzmāye] [shgāh] & Incorrect compound split \\
URL: https://... & [htt] [ps://...] & Non-linguistic oversegment \\
al-ketab & [al-ke] [tab] & Arabic morphology error \\
\bottomrule
\end{tabular}
|
29
|
2508.05628v1
| null |
\caption{Optimal hyperparameters determined through grid search on validation set}
|
\label{tab:hyper_sensitivity}
|
\begin{tabular}{lcc}
\toprule
\textbf{Hyperparameter} & \textbf{Range Tested} & \textbf{Optimal Value} \\
\midrule
Learning rate & $[5 \times 10^{-5}, 5 \times 10^{-4}]$ & $2 \times 10^{-4}$ \\
Warmup steps & $[10\text{k}, 100\text{k}]$ & $50\text{k}$ \\
Gumbel temperature & $[0.1, 10.0]$ & $5.0 \rightarrow 0.1$ \\
Morphology loss weight & $[0.0, 1.0]$ & $0.1$ \\
Chunk length penalty & $[0.0, 0.5]$ & $0.05$ \\
Gradient clipping & $[0.1, 5.0]$ & $1.0$ \\
\bottomrule
\end{tabular}
|
30
|
2508.05628v1
| null |
\caption{Optimal hyperparameters determined through grid search on validation set.}
| null |
\begin{tabular}{lcc}
\toprule
Hyperparameter & Range Tested & Optimal Value \\
\midrule
Learning rate & [5e-5, 5e-4] & 2e-4 \\
Warmup steps & [10k, 100k] & 50k \\
Gumbel temperature & [0.1, 10.0] & 5.0 → 0.1 \\
Morphology loss weight & [0.0, 1.0] & 0.1 \\
Chunk length penalty & [0.0, 0.5] & 0.05 \\
Gradient clipping & [0.1, 5.0] & 1.0 \\
\bottomrule
\end{tabular}
|
31
|
2508.05628v1
| null |
\caption{Detailed corpus statistics showing balanced distribution across splits.}
| null |
\begin{tabular}{@{}l@{\hspace{0.5cm}}r@{\hspace{0.5cm}}r@{\hspace{0.5cm}}r@{\hspace{0.4cm}}r@{}}
\toprule
Statistic & Training & Validation & Test & Total \\
\midrule
Total documents & 2.1M & 117K & 105K & 2.32M \\
Total tokens & 1.42B & 79M & 71M & 1.58B \\
Avg doc length (tokens) & 667 & 675 & 652 & 666 \\
Vocabulary size & 487K & 128K & 89K & 512K \\
\% with ZWNJ & 89.3\% & 88.9\% & 88.7\% & 89.2\% \\
\% with Latin script & 12.4\% & 11.8\% & 12.1\% & 12.3\% \\
\% with Arabic words & 8.7\% & 8.9\% & 8.5\% & 8.7\% \\
\bottomrule
\end{tabular}
|
14,523
|
2501.16885
| null | null | null | null |
32
|
2508.05628v1
| null |
\caption{Distribution of ZWNJ usage contexts in the training corpus.}
| null |
\begin{tabular}{lcc}
\toprule
ZWNJ Context & Frequency & Example \\
\midrule
Compound words & 45.2\% & dānesh-gāh (university) \\
Verbal constructions & 28.7\% & mi-ravam (I go) \\
Plural markers & 15.3\% & ketāb-hā(books) \\
Affixes & 7.8\% & nā-omid (hopeless) \\
Other & 3.0\% & -- \\
\bottomrule
\end{tabular}
|
33
|
2508.00311
| null |
\caption{Statistics of the dataset at line, paragraph, and page levels, showing the number of samples before deduplication as well as the split between the training and test sets after deduplication.}
|
\label{tab:dedup}
|
\begin{tabular}{l|c|c|c}
\hline
\textbf{Level} & \textbf{Before Deduplication} & \textbf{Train} & \textbf{Test} \\
\hline
Line-level & 1,884,532 & 741,016 & 1,000 \\
Paragraph-level & 412,789 & 135,575 & 1,000 \\
Page-level & 412,789 & 131,876 & 1,000 \\
\hline
\end{tabular}
|
34
|
2508.00311
| null |
\caption{
Comparison of formula recognition performance on three benchmarks.
All metrics are ED, lower is better.
The best results for each column are highlighted in \textbf{bold}, and the second best are \underline{underlined}.
}
|
\label{tab:ed_model_comparison}
|
\begin{tabular}{l l c cccc ccccc}
\toprule
\textbf{Model Type} & \textbf{Methods}
& \multicolumn{1}{c}{\textbf{Im2LaTeX-160k}}
& \multicolumn{5}{c}{\textbf{UniMER}}
& \multicolumn{4}{c}{\textbf{CSFormula }} \\
\cmidrule(lr){3-3} \cmidrule(lr){4-8} \cmidrule(lr){9-12}
& & SPE$\downarrow$
& SPE$\downarrow$ & CPE$\downarrow$ & SCE$\downarrow$ & HWE$\downarrow$ & Avg.$\downarrow$
& Line$\downarrow$ & Paragraph$\downarrow$ & Page$\downarrow$ & Avg.$\downarrow$ \\
\midrule
Task-Specific (SOTA) & UniMERNet & \textbf{0.240} & \textbf{0.060} & \textbf{0.056} & \underline{0.224} & \underline{0.072} & \underline{0.103} & 0.489 & 0.645 & 0.903 & 0.679 \\
Open VLM (SOTA) & Qwen2.5-VL & 0.310 & 0.185 & 0.460 & 0.364 & 0.206 & 0.303 & 0.416 & \underline{0.315} & 0.684 & 0.472 \\
Close VLM (SOTA) & GPT-4o & 0.434 & 0.497 & 0.528 & 0.644 & 0.512 & 0.545 & 0.338 & 0.357 & 0.511 & 0.402\\
Close VLM (SOTA) & Gemini-2.5-flash & 0.424 & 0.481 & 0.571 & 0.601 & 0.472 & 0.531 & \underline{0.312} & 0.378 & \underline{0.494} & \underline{0.394} \\
Professional Tool (SOTA) & Mathpix & 0.449 & 0.467 & 0.474 & 0.589 & 0.535 & 0.516 & 0.407 & 0.446 & 0.518 & 0.457 \\
\hline
& DocTron-Formula & \underline{0.245} & \underline{0.081} & \underline{0.084} & \textbf{0.182} & \textbf{0.046} & \textbf{0.098} & \textbf{0.121} & \textbf{0.121} & \textbf{0.251} & \textbf{0.164} \\
\bottomrule
\end{tabular}
|
35
|
2508.00311
| null |
\caption{
Comparison of formula recognition performance on three benchmarks.
All metrics are CDM, higher is better.
The best results for each column are highlighted in \textbf{bold}, and the second best are \underline{underlined}.
}
|
\label{tab:cdm_model_comparison}
|
\begin{tabular}{l l c ccccc ccccc}
\toprule
\textbf{Model Type} & \textbf{Method}
& \multicolumn{1}{c}{\textbf{Im2LaTeX-160k}}
& \multicolumn{5}{c}{\textbf{UniMER}}
& \multicolumn{4}{c}{\textbf{CSFormula}} \\
\cmidrule(lr){3-3} \cmidrule(lr){4-8} \cmidrule(lr){9-12}
& & SPE$\uparrow$
& SPE$\uparrow$ & CPE$\uparrow$ & SCE$\uparrow$ & HWE$\uparrow$ & Avg.$\uparrow$
& Line$\uparrow$ & Paragraph$\uparrow$ & Page$\uparrow$ & Avg.$\uparrow$
\\
\midrule
Task-Specific (SOTA) & UniMERNet & \textbf{0.991} & \textbf{0.994} & \textbf{0.970} & 0.946 & \textbf{0.953} & \textbf{0.965} & 0.919 & 0.644 & 0.009 & 0.524 \\
Open VLM (SOTA) & Qwen2.5-VL & 0.971 & 0.952 & 0.818 & \underline{0.947} & 0.927 & 0.911 & {0.924} & \underline{0.746} & 0.197 & 0.622 \\
Close VLM (SOTA) & GPT4o & 0.929 & 0.786 & 0.641 & 0.866 & 0.842 & 0.783 & 0.879 & 0.569 & 0.161 & 0.536 \\
Close VLM (SOTA) & Gemini-2.5-flash & {0.973} & {0.965} & 0.739 & 0.929 & 0.898 & 0.882 & 0.880 & {0.725} & \underline{0.592} & {0.732} \\
Professional Tool (SOTA) & Mathpix & 0.969 & 0.973 & \underline{0.967} & 0.932 & 0.924 & 0.949 & \underline{0.926} & 0.696 & {0.579} & \underline{0.733} \\
\hline
& DocTron-Formula & \underline{0.985} & \underline{0.979} & 0.962 & \textbf{0.958} & \underline{0.947} & \underline{0.961} & \textbf{0.950} & \textbf{0.897} & \textbf{0.774} & \textbf{0.873} \\
\bottomrule
\end{tabular}
|
36
|
2508.00311
| null |
\caption{Ablation study on the impact of different training data levels for DocTron-Formula.}
|
\label{tab:ablation_multilevel}
|
\begin{tabular}{l ccc ccc}
\toprule
\textbf{Exp}
& \multicolumn{3}{c}{\textbf{Training Data}}
& \multicolumn{3}{c}{\textbf{Testing Data}} \\
\cmidrule(lr){2-4} \cmidrule(lr){5-7}
& Line-level & Paragraph-level & Page-level
& Line-level $\downarrow$ & Paragraph-level $\downarrow$ & Page-level $\downarrow$ \\
\midrule
1 & \Checkmark & {\color[gray]{0.8} \ding{55}} & {\color[gray]{0.8} \ding{55}} & 0.123 & -- & -- \\
2 & {\color[gray]{0.8} \ding{55}} & \Checkmark & {\color[gray]{0.8} \ding{55}} & -- & 0.137 & -- \\
3 & {\color[gray]{0.8} \ding{55}} & {\color[gray]{0.8} \ding{55}} & \Checkmark & -- & -- & 0.278 \\
4 & \Checkmark & \Checkmark & {\color[gray]{0.8} \ding{55}} & 0.122 & 0.126 & -- \\
5 & \Checkmark & \Checkmark & \Checkmark & \textbf{0.121} & \textbf{0.123} & \textbf{0.272} \\
\bottomrule
\end{tabular}
|
37
|
2508.00311
| null |
\caption{Comparison of different model sizes and methods on three benchmarks. All metrics are ED, lower is better.}
|
\label{tab:model_size_comparison}
|
\begin{tabular}{l c ccccc ccccc}
\toprule
\textbf{Model} & \textbf{Size}
& \multicolumn{1}{c}{\textbf{Im2LaTeX-160k}}
& \multicolumn{5}{c}{\textbf{UniMER-Test}}
& \multicolumn{4}{c}{\textbf{CSFormula}}
\\
\cmidrule(lr){3-3} \cmidrule(lr){4-8} \cmidrule(lr){9-12}
& & SPE$\downarrow$
& SPE$\downarrow$ & CPE$\downarrow$ & SCE$\downarrow$ & HWE$\downarrow$ & Avg.$\downarrow$
& Line$\downarrow$ & Paragraph$\downarrow$ & Page$\downarrow$ & Avg.$\downarrow$
\\
\midrule
Qwen2.5-VL & 3B & 0.322 & 0.190 & \textbf{0.235} & 0.436 & 0.226 & \textbf{0.271} & 0.475 & 0.539 & 0.843 & 0.619 \\
Qwen2.5-VL & 7B & \textbf{0.310} & \textbf{0.185} & 0.460 & \textbf{0.364} & \textbf{0.206} & {0.303} & \textbf{0.416} & \textbf{0.315} & \textbf{0.684} & \textbf{0.472} \\
\midrule
DocTron-Formula & 3B & 0.253 & 0.095 & 0.099 & \textbf{0.177} & 0.055 & 0.106 & 0.122 & 0.131 & 0.297 & 0.183 \\
DocTron-Formula & 7B & \textbf{0.245} & \textbf{0.081} & \textbf{0.084} & 0.182 & \textbf{0.046} & \textbf{0.098} & \textbf{0.121} & \textbf{0.121} & \textbf{0.251} & \textbf{0.164} \\
\bottomrule
\end{tabular}
|
38
|
2508.00539
| null |
\caption{Comparison of filtering methods on Cuprite HSI (150-band subset).}
|
\label{tab:filter_comparison}
|
\begin{tabular}{lcccc}
\toprule
\multirow{2}{*}{Method} & \multicolumn{2}{c}{Full Bands} & \multicolumn{2}{c}{Selected Bands} \\
\cmidrule(lr){2-3} \cmidrule(lr){4-5}
& CosSim & RMSE & CosSim & SNR \\
\midrule
Fourier Transform & 0.9520 & 0.5669 & 0.8649 & 0.5305 \\
Wavelet Transform & 0.8757 & 0.5771 & 0.9295 & 0.5049 \\
Phase-Locked Filtering & 0.8932 & 0.8932 & \textbf{0.9952} & \textbf{0.4502} \\
\bottomrule
\end{tabular}
|
18,978
|
2501.19195
| null | null |
\label{proposition:OptimalErrorRate}
| null |
39
|
2508.00476
| null |
\caption{Mean and standard deviation for LLM-as-judge evaluation on Czech Answers.}
|
\label{tab:mean_std_cz}
|
\begin{tabular}{|l|r|}
\hline
Model & Mean $\pm$ 1 std \\
\hline
GETALP@AutoMin & 5.15 $\pm$ 3.73 \\
GETALP@AutoMin\_amr & 4.97 $\pm$ 3.77 \\
GETALP@AutoMin\_amr\_only & 4.31 $\pm$ 3.52\\
\hline
\end{tabular}
|
40
|
2508.00476
| null |
\caption{Mean and standard deviation for LLM-as-judge and Human evaluation on English Answers.}
|
\label{tab:mean_std_en}
|
\begin{tabular}{|l|r|}
\hline
Model & Mean $\pm$ 1 std \\
\hline
\multicolumn{2}{c}{\textbf{LLM-as-Judge}} \\
\hline
GETALP@AutoMin & 4.09 $\pm$ 3.16 \\
GETALP@AutoMin\_amr & 3.35 $\pm$ 2.54 \\
GETALP@AutoMin\_amr\_only & 2.46 $\pm$ 1.75 \\
\hline
\multicolumn{2}{c}{\textbf{Human evaluation}} \\
\hline
GETALP@AutoMin & 5.65 $\pm$ 3.06 \\
GETALP@AutoMin\_amr & 5.55 $\pm$ 2.95 \\
GETALP@AutoMin\_amr\_only & 3.94 $\pm$ 2.69\\
\hline
\end{tabular}
|
41
|
2508.00476
| null |
\caption{Examples where LLM-as-judge gave a score of 2 while evaluators gave a score of either 8 or 10, for both GETALP@Automin and GETALP@Automin\_amr models. The answer displayed comes from one of the two models.}
|
\label{tab:manual_good}
|
\begin{tabular}{|c|c|}
\hline
Ground Truth & Ours \\
\hline
\multicolumn{2}{|c|}{\textbf{Who is leaving the project?}} \\
\hline
[PERSON10] & [PERSON10] will be leaving. \\
\hline
\multicolumn{2}{|c|}{\textbf{What is the application deadline for the PhD program?}} \\
\hline
End of the year & The PhD application deadline is the end of the year. \\
\hline
\multicolumn{2}{|c|}{\textbf{Who had to pause the meeting for a few minutes?}} \\
\hline
[PERSON2] & PERSON2 had to pause the meeting for a few minutes. \\
\hline
\end{tabular}
|
42
|
2508.00476
| null |
\caption{Examples where both LLM-as-judge and evaluators gave a score 2, for both GETALP@Automin and GETALP@Automin\_amr models. The answer displayed comes from one of the two models.}
|
\label{tab:manual_bad}
|
\begin{tabular}{|c|c|}
\hline
Ground Truth & Ours \\
\hline
\multicolumn{2}{|c|}{\textbf{What is [PERSON4] focusing on now?}} \\
\hline
Multi-source machine translation & [PERSON4] is putting together and also working on the censorship \\ & component, which is one part of the pipeline after the [PROJECT4]. \\
\hline
\multicolumn{2}{|c|}{\textbf{What is the current delay of the language id system?}} \\
\hline
3 seconds & The current delay of the language ID system is not explicitly stated.\\
\hline
\multicolumn{2}{|c|}{\textbf{How did [PERSON7] qualify the experience of the latest recording session assessment?}} \\
\hline
It was qualified as "disastrous". & Unfortunately, the provided information does not contain a direct \\ & quote from [PERSON7] regarding the experience of the latest recording \\ & session assessment. \\
\hline
\end{tabular}
|
43
|
2508.00721
| null |
\caption{A comparison between foundation, domain-specific, and untrained priors for Gaussian deblurring on AFHQ (resolution: $256 \times 256$; DS: domain-specific; FD: foundation). \textbf{Bold}: best, \& \underbar{underline}: second best, for each metric/column}
|
\label{tab:generic_vs_domain}
|
\begin{tabular}{c c c c c}
\hline
% &\multicolumn{4}{c}{\scriptsize{\textbf{Gaussian Deblur ($256 \times 256$)}}}
% \\
% \cmidrule(lr){2-5}
&\scriptsize{PSNR$\uparrow$}
&\scriptsize{SSIM$\uparrow$}
&\scriptsize{LPIPS$\downarrow$}
&\scriptsize{CLIPIQA$\uparrow$}
\\
\hline
\scriptsize{\textbf{DIP}}
&\scriptsize{\textbf{32.9470}}
&\scriptsize{\textbf{0.9075}}
&\scriptsize{\underbar{0.1873}}
&\scriptsize{\underbar{0.5024}}
\\
\hline
\scriptsize{\textbf{D-Flow (DS)}}
&\scriptsize{\underbar{30.2512}}
&\scriptsize{\underbar{0.8553}}
&\scriptsize{\textbf{0.1183}}
&\scriptsize{\textbf{0.6536}}
\\
\hline
\scriptsize{\textbf{D-Flow (FD)}}
&\scriptsize{30.0975}
&\scriptsize{0.8861}
&\scriptsize{0.2270}
&\scriptsize{0.5404}
\\
\hline
\end{tabular}
|
44
|
2508.00721
| null |
\caption{\textbf{x4 Super Resolution} with additive Gaussian noise ($\sigma = 0.03$). (\textbf{Bold}: best, \underbar{underline}: second best)}
|
\label{tab:sr4}
|
\begin{tabular}{c c c c c c c c c}
\hline
&\multicolumn{4}{c}{\scriptsize{\textbf{DIV2K~\cite{Agustsson_2017_CVPR_Workshops}}}}
&\multicolumn{4}{c}{\scriptsize{\textbf{AFHQ~\cite{choi2020starganv2}}}}
\\
\cmidrule(lr){2-5}
\cmidrule(lr){6-9}
&\scriptsize{PSNR$\uparrow$}
&\scriptsize{SSIM$\uparrow$}
&\scriptsize{LPIPS$\downarrow$}
&\scriptsize{CLIPIQA$\uparrow$}
&\scriptsize{PSNR$\uparrow$}
&\scriptsize{SSIM$\uparrow$}
&\scriptsize{LPIPS$\downarrow$}
&\scriptsize{CLIPIQA$\uparrow$}
\\
\hline
\scriptsize{\textbf{DIP}}
&\scriptsize{24.896}
&\scriptsize{0.613}
&\scriptsize{0.482}
&\scriptsize{0.305}
&\scriptsize{28.275}
&\scriptsize{0.697}
&\scriptsize{0.466}
&\scriptsize{0.284}
\\
\hline
\scriptsize{\textbf{FlowChef}}
&\scriptsize{24.439}
&\scriptsize{0.677}
&\scriptsize{0.444}
&\scriptsize{\textbf{0.630}}
&\scriptsize{28.444}
&\scriptsize{0.779}
&\scriptsize{0.404}
&\scriptsize{\textbf{0.568}}
\\
\hline
\scriptsize{\textbf{FlowDPS}}
&\scriptsize{24.284}
&\scriptsize{0.657}
&\scriptsize{0.461}
&\scriptsize{0.432}
&\scriptsize{28.507}
&\scriptsize{0.743}
&\scriptsize{0.445}
&\scriptsize{0.293}
\\
\hline
\scriptsize{\textbf{D-Flow}}
% step 3 res
&\scriptsize{24.467}
&\scriptsize{0.684}
&\scriptsize{0.488}
&\scriptsize{0.387}
&\scriptsize{27.596}
&\scriptsize{0.709}
&\scriptsize{0.518}
&\scriptsize{0.296}
\\
\hline
\scriptsize{\textbf{FMPlug-W}}
% step 3 res
&\scriptsize{\underbar{26.047}}
&\scriptsize{\underbar{0.762}}
&\scriptsize{\underbar{0.369}}
&\scriptsize{0.410}
&\scriptsize{\underbar{30.234}}
&\scriptsize{\underbar{0.813}}
&\scriptsize{\underbar{0.358}}
&\scriptsize{0.367}
\\
\hline
\scriptsize{\textbf{FMPlug-W-R}}
% step 3 res
&\scriptsize{\textbf{26.144}}
&\scriptsize{\textbf{0.769}}
&\scriptsize{\textbf{0.355}}
&\scriptsize{\underbar{0.497}}
&\scriptsize{\textbf{30.449}}
&\scriptsize{\textbf{0.821}}
&\scriptsize{\textbf{0.340}}
&\scriptsize{\underbar{0.473}}
\\
\hline
\end{tabular}
|
45
|
2508.00721
| null |
\caption{\textbf{Gaussian Deblur} with additive Gaussian noise ($\sigma = 0.03$). (\textbf{Bold}: best, \underbar{underline}: second best)}
|
\label{tab:gaussian}
|
\begin{tabular}{c c c c c c c c c}
\hline
&\multicolumn{4}{c}{\scriptsize{\textbf{DIV2K~\cite{Agustsson_2017_CVPR_Workshops}}}}
&\multicolumn{4}{c}{\scriptsize{\textbf{AFHQ~\cite{choi2020starganv2}}}}
\\
\cmidrule(lr){2-5}
\cmidrule(lr){6-9}
&\scriptsize{PSNR$\uparrow$}
&\scriptsize{SSIM$\uparrow$}
&\scriptsize{LPIPS$\downarrow$}
&\scriptsize{CLIPIQA$\uparrow$}
&\scriptsize{PSNR$\uparrow$}
&\scriptsize{SSIM$\uparrow$}
&\scriptsize{LPIPS$\downarrow$}
&\scriptsize{CLIPIQA$\uparrow$}
\\
\hline
\scriptsize{\textbf{DIP}}
&\scriptsize{24.659}
&\scriptsize{0.658}
&\scriptsize{0.469}
&\scriptsize{0.367}
&\scriptsize{28.169}
&\scriptsize{0.680}
&\scriptsize{0.472}
&\scriptsize{0.283}
\\
\hline
\scriptsize{\textbf{FlowChef}}
&\scriptsize{20.248}
&\scriptsize{0.474}
&\scriptsize{0.629}
&\scriptsize{0.236}
&\scriptsize{24.366}
&\scriptsize{0.661}
&\scriptsize{0.533}
&\scriptsize{\underbar{0.298}}
\\
\hline
\scriptsize{\textbf{FlowDPS}}
&\scriptsize{19.900}
&\scriptsize{0.436}
&\scriptsize{0.615}
&\scriptsize{0.190}
&\scriptsize{24.612}
&\scriptsize{0.626}
&\scriptsize{0.532}
&\scriptsize{0.155}
\\
\hline
\scriptsize{\textbf{D-Flow}}
% step 3 res
&\scriptsize{24.798}
&\scriptsize{0.683}
&\scriptsize{0.390}
&\scriptsize{\textbf{0.464}}
&\scriptsize{28.955}
&\scriptsize{0.752}
&\scriptsize{0.458}
&\scriptsize{\textbf{0.316}}
\\
\hline
\scriptsize{\textbf{FMPlug-W}}
&\scriptsize{\underbar{25.995}}
&\scriptsize{\underbar{0.749}}
&\scriptsize{\underbar{0.387}}
&\scriptsize{\underbar{0.413}}
&\scriptsize{\underbar{30.315}}
&\scriptsize{\underbar{0.805}}
&\scriptsize{\underbar{0.378}}
&\scriptsize{0.246}
\\
\hline
\scriptsize{\textbf{FMPlug-W-R}}
% step 3 res
&\scriptsize{\textbf{26.125}}
&\scriptsize{\textbf{0.757}}
&\scriptsize{\textbf{0.378}}
&\scriptsize{0.412}
&\scriptsize{\textbf{30.322}}
&\scriptsize{\textbf{0.808}}
&\scriptsize{\textbf{0.372}}
&\scriptsize{0.248}
\\
\hline
\end{tabular}
|
46
|
2508.00700
| null |
\caption{Distribution of Severities Per Quality Metric. }
|
\label{tab:severities}
|
\begin{tabular}{cc|l|r|r|}
\toprule
\multicolumn{2}{|c|}{\textbf{Dataset}} &
\multicolumn{1}{c|}{\textbf{\begin{tabular}[c]{@{}c@{}}Severity \\ Type\end{tabular}}} &
\multicolumn{1}{c|}{\textbf{\begin{tabular}[c]{@{}c@{}}Maintainability\\ Code Smell(\%)\end{tabular}}} &
\multicolumn{1}{c|}{\textbf{\begin{tabular}[c]{@{}c@{}}Reliability\\Bugs(\%)\end{tabular}}} \\ \midrule
\multicolumn{1}{|c|}{\multirow{14}{*}{Introductory}} & \multirow{4}{*}{\rotatebox[origin=c]{90}{Human}} & BLOCKER & 0 & 87.50 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & CRITICAL & 2.58 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 9.64 & 1.56 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 9.81 & 0 \\ \cmidrule(l){2-5}
\multicolumn{1}{|c|}{} & \multirow{4}{*}{\rotatebox[origin=c]{90}{Zero-Shot}} & BLOCKER & 0.06 & 3.13 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & CRITICAL & 2.58 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 16 & 4.69 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 32 & 0 \\ \cmidrule(l){2-5}
\multicolumn{1}{|c|}{} & \multirow{4}{*}{\rotatebox[origin=c]{90}{Few-Shot}} & BLOCKER & 0 & 1.56 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & CRITICAL & 3 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 11 & 1.56 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 13 & 0 \\ \cmidrule(l){2-5}
\multicolumn{1}{|c|}{} & \multirow{2}{*}{\rotatebox[origin=c]{90}{Fine-T}} & MINOR & 0.45 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & All Others & 0 & 0 \\ \midrule
\multicolumn{1}{|c|}{\multirow{16}{*}{Interview}} & \multirow{4}{*}{\rotatebox[origin=c]{90}{Human}} & BLOCKER & 0.05 & 71.27 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & CRITICAL & 3.35 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 9.82 & 11.33 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 14.71 & 0.55 \\ \cmidrule(l){2-5}
\multicolumn{1}{|c|}{} & \multirow{4}{*}{\rotatebox[origin=c]{90}{Zero-Shot}} & BLOCKER & 0.03 & 0.28 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & CRITICAL & 4.94 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 15.04 & 3.04 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 13.08 & 0.55 \\ \cmidrule(l){2-5}
\multicolumn{1}{|c|}{} & \multirow{4}{*}{\rotatebox[origin=c]{90}{Few-Shot}} & BLOCKER & 0 & 0.55 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & CRITICAL & 4.64 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 10.15 & 1.93 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 8.94 & 1.38 \\ \cmidrule(l){2-5}
\multicolumn{1}{|c|}{} & \multirow{4}{*}{\rotatebox[origin=c]{90}{Fine-T}} & BLOCKER & 0.01 & 8.56 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & CRITICAL & 0.18 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 0.30 & 0.55 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 14.76 & 0 \\ \midrule
\multicolumn{1}{|c|}{\multirow{14}{*}{Competition}} & \multirow{4}{*}{\rotatebox[origin=c]{90}{Human}} & BLOCKER & 0.11 & 86.30 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & CRITICAL & 4.96 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 15.86 & 8.22 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 19.49 & 0 \\ \cmidrule(l){2-5}
\multicolumn{1}{|c|}{} & \multirow{4}{*}{\rotatebox[origin=c]{90}{Zero-Shot}} & BLOCKER & 0.11 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & CRITICAL & 7.38 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 12.67 & 1.37 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 11.67 & 0 \\ \cmidrule(l){2-5}
\multicolumn{1}{|c|}{} & \multirow{3}{*}{\rotatebox[origin=c]{90}{Few-Shot}} & CRITICAL & 7.16 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 10.68 & 1.37 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 8.92 & 1.37 \\ \cmidrule(l){2-5}
\multicolumn{1}{|c|}{} & \multirow{3}{*}{\rotatebox[origin=c]{90}{Fine-T}} & CRITICAL & 0.33 & 0 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MAJOR & 0.22 & 1.37 \\ %\cmidrule(l){3-5}
\multicolumn{1}{|c|}{} & & MINOR & 0.44 & 0 \\ \bottomrule
\end{tabular}
|
47
|
2508.00700
| null |
\caption{Odds Ratio for Reliability (Bugs).}
|
\label{tab:stat-odds}
|
\begin{tabular}{llll}
\toprule
\multicolumn{1}{l|}{Test} & introductory & interview & competition \\
\midrule
\multicolumn{1}{l|}{Zero-Shot x Human} & 0.008*** & 0.006*** & 0.0005*** \\
\multicolumn{1}{l|}{Few-Shot x Human } & 0.003*** & 0.006*** & 0.001*** \\
\multicolumn{1}{l|}{Fine-Tuning x Human } & 0.00*** & 0.019*** & 0.0005*** \\
\multicolumn{1}{l|}{Few-Shot x Zero-Shot } & 0.410 & 1.000 & 2.020 \\
\multicolumn{1}{l|}{Fine-Tuning x Zero-Shot} & 0.000** & 3.200 & 1.000 \\
\bottomrule
\multicolumn{2}{l}{p\textless{}0.05* p\textless{}0.01** p\textless{}0.001***} & &
\end{tabular}
|
20,109
|
2502.00298
| null | null |
\label{lemma:test-train-bound}
| null |
48
|
2508.00700
| null |
\caption{Odds Ratio for Maintainability (Code Smells).}
|
\label{tab:stat-smells}
|
\begin{tabular}{llll}
\toprule
\multicolumn{1}{l|}{Test} & introductory & interview & competition \\
\midrule
\multicolumn{1}{l|}{Zero-Shot x Human} & 3.57*** & 1.3 & 0.67 \\
\multicolumn{1}{l|}{Few-Shot x Human } & 1.29 & 0.80 & 0.52* \\
\multicolumn{1}{l|}{Fine-Tuning x Human } & 0.00*** & 0.47 & 0.00*** \\
\multicolumn{1}{l|}{Few-Shot x Zero-Shot } & 0.36* & 0.60 & 0.78 \\
\multicolumn{1}{l|}{Fine-Tuning x Zero-Shot} & 0.00*** & 0.35** & 0.00*** \\
\bottomrule
\multicolumn{2}{l}{p\textless{}0.05* p\textless{}0.01** p\textless{}0.001*** } & &
\end{tabular}
|
49
|
2508.00700
| null |
\caption{Mann-Whitney and Cliff's Delta for Effort Time.}
|
\label{tab:stat-efforttime}
|
\begin{tabular}{llll}
\toprule
\multicolumn{1}{l|}{Test} & introductory & interview & competition \\ \midrule
\multicolumn{1}{l|}{ZS x H} & 0.54 large*** & 0.21 small*** & 0.1 negligible* \\
\multicolumn{1}{l|}{FS X H} & 0.38 medium*** & 0.14 negligible*** & 0.007 negligible \\
\multicolumn{1}{l|}{FT x H} & 0.24 small & 0.27 small*** & -0.08 small \\
\multicolumn{1}{l|}{FS x ZS} & -0.19 small*** & -0.08 negligible*** & -0.09 negligible* \\
\multicolumn{1}{l|}{FT x ZS} & -0.50 large & 0.03 negligible & -0.22 small \\
\bottomrule
\multicolumn{4}{l}{Human(H), Zero-Shot(ZS), Few-Shot(FS), Fine-Tuning(FT)} \\
\multicolumn{4}{l}{p\textless{}0.05* p\textless{}0.01** p\textless{}0.001***}
\end{tabular}
|
50
|
2508.00700
| null |
\caption{Pass@1 results.}
|
\label{tab:correctnesstable}
|
\begin{tabular}{l|c|c|c|c}
\toprule
Test & Human & Zero-Shot & Few-Shot & Fine-Tuned \\
\midrule
Pass@1 & 1.0 & 0.53 & 0.87 & 0.47 \\
\bottomrule
\end{tabular}
|
51
|
2508.00288
| null |
\caption{Comparison of ObjectNav and Aerial Navigation benchmarks.}
|
\label{tab:dataset_comparison}
|
\begin{tabular}{ccccc}
\toprule
\textbf{Benchmark} & \textbf{Viewpoint} & \textbf{Task} & \textbf{Goal Type} & \textbf{Goal Specification} \\
\midrule
AI2-THOR~\cite{kolve2017ai2thor} & Ground & ObjNav & Category & Category Label \\
Gibson~\cite{xia2018gibson} & Ground & ObjNav & Category & Category Label \\
RoboTHOR~\cite{deitke2020robothor} & Ground & ObjNav & Category & Category Label \\
HM3D~\cite{ramakrishnan2021hm3d} & Ground & ObjNav & Category & Category Label \\
GeoText~\cite{chu2024geotext1652} & Aerial & VLN & Location & Movement Instruction \\
AerialVLN~\cite{liu2023aerialvln} & Aerial & VLN & Location & Movement Instruction \\
CityNav~\cite{lee2024citynav} & Aerial & VLN & Location & Movement Instruction \\
TravelUAV~\cite{wang2024} & Aerial & VLN & Location & Movement Instruction \\
OpenFly~\cite{gao2025} & Aerial & VLN & Location & Movement Instruction \\
\textbf{UAV-ON (Ours)} & \textbf{Aerial} & \textbf{ObjNav} & \textbf{Instance} & \textbf{Semantic Instruction} \\
\bottomrule
\end{tabular}
|
52
|
2508.00288
| null |
\caption{Performance across different object sizes and overall total, evaluated using four metrics: Distance to Success (DTS↓, in simulation units), Success Rate (SR↑), Oracle Success Rate (OSR↑), and Success weighted by Path Length (SPL↑).}
|
\label{tab:size_results}
|
\begin{tabular}{lcccccccccccccccc}
\toprule
\multirow{2}{*}{\textbf{Method}}
& \multicolumn{4}{c}{\textbf{Small}}
& \multicolumn{4}{c}{\textbf{Medium}}
& \multicolumn{4}{c}{\textbf{Large}}
& \multicolumn{4}{c}{\textbf{Total}} \\
% \cmidrule(lr){2-5} \cmidrule(lr){6-9} \cmidrule(lr){10-13} \cmidrule(lr){14-17}
& DTS↓ & SR↑ & OSR↑ & SPL↑
& DTS↓ & SR↑ & OSR↑ & SPL↑
& DTS↓ & SR↑ & OSR↑ & SPL↑
& DTS↓ & SR↑ & OSR↑ & SPL↑ \\
\midrule
Random
& \textbf{43.08} & 4.14\% & 7.80\% & \textbf{2.80}\%
& \textbf{42.62} & 3.33\% & 8.10\% & 3.05\%
& 40.59 & 2.48\% & 8.07\% & 1.62\%
& \textbf{42.57} & 3.70\% & 8.00\% & 2.66\% \\
CLIP-H
& 48.64 & 2.86\% & 8.43\% & 1.51\%
& 43.91 & \textbf{10.95}\% & 16.67\% & \textbf{7.17}\%
& 40.38 & 13.04\% & 19.25\% & 10.53\%
& 46.31 & 6.20\% & 11.90\% & \textbf{4.15}\% \\
AOA-V
& 50.56 & 2.86\% & \textbf{25.44}\% & 0.54\%
& 48.96 & 5.71\% & \textbf{27.62}\% & 1.73\%
& 46.58 & 7.45\% & \textbf{27.95}\% & 1.038\%
& 49.58 & 4.20\% & \textbf{26.30}\% & 0.87\% \\
AOA-F
& 50.67 & \textbf{4.45}\% & 16.38\% &1.61\%
& 48.14 & 10.48\% & 17.62\% & 6.36\%
& \textbf{39.68} & 14.29\% & 21.74\% & \textbf{10.66}\%
& 48.37 & \textbf{7.30}\% & 17.50\% & 4.06\% \\
\bottomrule
\end{tabular}
|
53
|
2508.00288
| null | null |
\label{tab:termination_distance}
|
\begin{tabular}{lcccccc}
\toprule
\multirow{2}{*}{\textbf{Method}}
& \multicolumn{3}{c}{\textbf{Termination Type (\%)}}
& \multicolumn{2}{c}{\textbf{Navigation Statistics}} \\
& Stop & Max Step & Collision & Avg. Steps & Safe Dist. (m) \\
\midrule
Random & \textbf{62.1} & 0.0 & 37.9 & 12.09 & 31.68 \\
CLIP-H & 36.8 & 11.8 & 51.4 & 30.25 & 125.21 \\
AOA-V & 19.9 & \textbf{35.1} & 45.0 & \textbf{85.65} & \textbf{232.08} \\
AOA-F & 30.6 & 3.9 & \textbf{65.5} & 36.88 & 144.25 \\
\bottomrule
\end{tabular}
|
54
|
2508.00587
| null |
\caption{\textbf{Quantitative results.} Comparison with the state-of-the-art on the official SMIYC~\cite{chan2021segmentmeifyoucan} benchmark, Road Anomaly~\cite{lis2019detecting} and Fishyscapes Static~\cite{blum2021fishyscapes} validation set, along with the average performance across all five benchmark datasets. The best result for each dataset is highlighted in \textbf{bold}, and the second-best is \underline{underlined}. Notably, UEM~\cite{nayal2024likelihood} and PixOOD~\cite{vojivr2024pixood} also use a DINOv2 backbone.}
|
\label{tab:exp:quantitative}
|
\begin{tabular}{lrrrrrrrrrr|rr}
\toprule
\multirow{2}{*}{Method} & \multicolumn{2}{c}{Anomaly Track} & \multicolumn{2}{c}{Obstacle Track} & \multicolumn{2}{c}{LaF NoKnown} & \multicolumn{2}{c}{Road Anomaly} & \multicolumn{2}{c}{FS Static} & \multicolumn{2}{c}{Average}\\
%\cmidrule(lr){3-5}
%\cmidrule(lr){6-8}
%\cmidrule(lr){9-11}
\cmidrule(lr){2-3}
\cmidrule(lr){4-5}
\cmidrule(lr){6-7}
\cmidrule(lr){8-9}
\cmidrule(lr){10-11}
\cmidrule(lr){12-13}
&
\multicolumn{1}{c}{AP $\uparrow$}& \multicolumn{1}{c}{FPR $\downarrow$}&
\multicolumn{1}{c}{AP $\uparrow$}& \multicolumn{1}{c}{FPR $\downarrow$}&
\multicolumn{1}{c}{AP $\uparrow$}& \multicolumn{1}{c}{FPR $\downarrow$}&
\multicolumn{1}{c}{AP $\uparrow$}& \multicolumn{1}{c}{FPR $\downarrow$}&
\multicolumn{1}{c}{AP $\uparrow$}& \multicolumn{1}{c}{FPR $\downarrow$}&
\multicolumn{1}{c}{$\overline{\text{AP}}$ $\uparrow$}& \multicolumn{1}{c}{$\overline{\text{FPR}}$ $\downarrow$}\\
\midrule
SynBoost~\cite{di2021pixel} & 56.44 & 61.86 & 71.34 & 3.15 & 81.71 & 4.64 & -- & -- & 72.59 & 18.75 & 70.52 & 22.10\\
%ATTA~\cite{gao2023atta} & 67.04 & 31.57 & 76.46 & 2.81 & -- & -- \\
Maximized Entropy~\cite{chan2021entropy} & 85.47 & 15.00 & 85.07 & 0.75 & 77.90 & 9.70 & -- & -- & 81.00 & 5.00 & 82.36 & 7.61\\
PEBAL~\cite{tian2022pixel} & 49.14 & 40.82 & 4.98 & 12.68 & -- & -- & 62.37 & 28.29 & 82.73 & 6.81 & 49.81 & 22.15\\
DenseHybrid~\cite{grcic2022densehybrid} & 77.96 & 9.81 & 87.08 & 0.24 & 78.67 & 2.12 & -- & -- & -- & -- & 81.24 & 4.06 \\
DaCUP~\cite{vojivr2023image} & -- & -- & 81.50 & 1.13 & 81.37 & 7.36 & -- & -- & -- & -- & 81.44 & 4.25 \\
RbA~\cite{nayal2023rba} & 94.46 & 4.60 & \textbf{95.12} & \textbf{0.08} & -- & -- & 85.42 & 6.92 & -- & -- & \underline{91.67} & 3.87 \\
EAM~\cite{grcic2023advantages} & 93.75 & \underline{4.09} & 92.87 & 0.52 & -- & -- & 69.40 & 7.70 & \underline{96.00} & \underline{0.30} & 88.51 & \underline{3.15} \\
NFlowJS~\cite{grcic2024dense} & 56.92 & 34.71 & 85.55 & 50.36 & \textbf{89.28} & \textbf{0.65} & -- & -- & -- & -- & 77.25 & 28.57\\
Mask2Anomaly~\cite{rai2023unmasking} & 88.72 & 14.63 & 93.22 & 0.20 & -- & -- & 79.70 & 13.45 & 95.20 & 0.82 & 89.66 & 5.18\\
RPL\texttt{+}CoroCL~\cite{liu2023residual} & 83.49 & 11.68 & 85.93 & 0.58 & -- & -- & 71.60 & 17.74 & 92.46 & 0.85 & 83.87 & 7.71 \\
Maskomaly~\cite{ackermann2023maskomaly} & 93.35 & 6.87 & -- & -- & -- & -- & 70.90 & 11.90 & 69.50 & 14.40 & 77.92 & 11.06 \\
%ODIN~\cite{liang2018enhancing} & 33.06 & 71.68 & 22.12 & 15.28 & 52.93 & 30.04 \\
%JSRNet~\cite{Vojir_2021_ICCV} & 33.64 & 43.85 & 28.09 & 28.86 & 74.17 & 6.59 \\
%Image Resynthesis~\cite{Lis_2019_ICCV} & 52.28 & 25.93 & 37.71 & 4.70 & 57.08 & 8.82 \\
cDNP~\cite{galesso2023far} & 88.90 & 11.42 & -- & -- & -- & -- & 85.60 & 9.80 & -- & -- & 87.25 & 10.61 \\
Road Inpainting~\cite{lis2023detecting} & -- & -- & 54.14 & 47.12 & 82.93 & 35.75 & -- & -- & -- & -- & 68.54 & 41.44 \\
%ObsNet~\cite{besnier2021trigger} & 75.44 & 26.69 & -- & -- & -- & -- \\
CSL~\cite{zhang2024csl} & 80.08 & 7.16 & 87.10 & 0.67 & -- & -- & 61.38 & 43.80 & -- & -- & 76.19 & 17.88\\
PixOOD~\cite{vojivr2024pixood} & 68.88 & 54.33 & 88.90 & 0.30 & \underline{85.07} & 4.46 & \textbf{96.39} &\textbf{ 4.30} & -- & -- & 84.31 & 15.85\\
UNO~\cite{delic2024outlier} & \textbf{96.33} & \textbf{1.98} & 93.19 & 0.16 & -- & -- & 85.50 & 7.40 & \textbf{98.00} & \textbf{0.04} & \textbf{94.24} & 3.52\\
UEM~\cite{nayal2024likelihood} & \underline{95.60} & 4.70 & \underline{94.38} & \underline{0.10} & 81.04 & \underline{1.45} & 90.94 & 8.03 & -- & -- & 90.49 & 3.57 \\
%\pixood ({\bf Ours}) & 68.88 & 54.33 & 19.82 &\first 88.90 &\first 0.30 & \second 50.82 &\second 85.07 &\second 4.46 & 44.41\\
\cmidrule(lr){1-13}
\rowcolor{gray!10}
BCE Baseline & 92.64 & 7.74 & 86.58 & 0.53 & 73.67 & 6.63 & 91.26 & 5.9 & 92.92 & 1.23 & 87.41 & 4.41\\
\rowcolor{gray!10}
EDL (ours) & 94.19 & 5.82 & 91.07 & 0.19 & 80.90 & 1.20 & \underline{92.95} & \underline{4.81} & 95.44 & 0.49 & 90.91 & \textbf{2.50}\\
\bottomrule
\end{tabular}
|
55
|
2508.00607
| null |
\caption{Selection rules of the DDI (Eq. \eqref{multipol_SF_spherical}), written for the symmetrized fully-coupled basis states.}
|
\label{tab:Selection_rules_DDI}
|
\begin{tabular}{cc}
\hline \hline
Quantum number & Selection rule \\
\hline\hline
\multirow{2}{*}{$[p_1,p_2]$} &
$[\pm,\pm] \leftrightarrow [\mp,\mp]$ \\
& $[\pm,\mp] \leftrightarrow [\mp,\pm]$ \\ [0.3ex]
$[\Delta j_1,\Delta j_2]$ & $[\pm 1,\pm 1]$ or $[\pm 1,\mp 1]$ \\[0.3ex]
$\Delta j_{12}$ & $0^{(*)}$ or $\pm 1$ or $\pm 2$ \\[0.3ex]
$\Delta L$ & $0^{(*)}$ or $\pm 2$\\[0.3ex]
$\Delta J$ & 0 \\[0.3ex]
$\Delta M$ & 0 \\[0.3ex]
Parity & $\pm \leftrightarrow \pm$ \\[0.3ex]
Reflection & $\pm \leftrightarrow \pm$ \\[0.3ex]
Permutation & $\pm \leftrightarrow \pm$ \\[0.3ex]
\hline\hline
\multicolumn{2}{l}{$^*$ $\Delta X=0$ except $0 \leftrightarrow 0$}
\\
\end{tabular}
|
138
|
2508.00398
| null |
\caption{Inference time and memory usage across stylization networks (excluding 3D diffusion model for isolated analysis).}
|
\label{tab:complexity}
|
\begin{tabular}{lcc}\toprule
Method & seconds/frame & memory \\\midrule
DSU \cite{zhou2024drawingspinup} & 0.276 & 11.62 GB\\
DSU \cite{zhou2024drawingspinup} + FDED (Ours) & 0.282 & 11.91 GB\\ \midrule
USNet (Ours) & 0.108 & 4.81 GB\\
USNet + FDED (Ours) & 0.115 & 5.05 GB\\ \bottomrule
\end{tabular}
|
20,110
|
2502.00298
| null | null |
\label{lemma:ski-test-train-bound}
| null |
56
|
2508.00607
| null |
\caption{Selection rules of the electric dipole-dipole interaction, the one-photon and two-photon electric-dipole optical transitions, written for the symmetrized fully-coupled basis states.}
|
\label{tab:selection_rules_1photon}
|
\begin{tabular}{ccc}
\hline \hline
Quantum & One-photon & One-photon \\
numbers & $\pi$ transition & $\sigma^{\pm}$ transition \\
\hline\hline
$\Delta S$ & $0$ & $0$ \\ [0.3ex]
$\Delta \Lambda$ & $0, \pm 1$ & $0, \pm 1$ \\[0.3ex]
$\Delta \Omega$ & $0, \pm 1$ & $0, \pm 1$ \\[0.3ex]
$\Delta j^{\text{a}}$ & $0, \pm 1^{\text{b}}$ & $0, \pm 1$ \\[0.3ex]
$\Delta m$ & $0$ & $\pm 1 $ \\[0.3ex]
Parity & $\pm \leftrightarrow \mp$ & $\pm \leftrightarrow \mp$ \\[0.3ex]
\hline \hline
\multicolumn{3}{l}{ $^{\text{a}}$ In the case $\Lambda =0 \rightarrow \Lambda =0$ or $\Omega =0 \rightarrow \Omega =0$,}\\
\multicolumn{3}{l}{only $\Delta j=\pm 1$}\\
\multicolumn{3}{l}{ $^{\text{b}}$ $\Delta j = \pm 1$ for $m=m'=0$}\\
\end{tabular}
|
57
|
2508.00607
| null |
\caption{The weights $(c^\infty_1)^2$, $(c^\infty_2)^2$, $(c^\infty_3)^2$ of the asymptotic states in the entrance scattering channel, for different Rabi frequencies $\Omega_1$ at fixed $\Omega_2 = 2\pi\times 200$~MHz.}
|
\label{strength_coupling}
|
\begin{tabular}{ccccc}
\hline \hline
$\Omega_1/2\pi$ (MHz) & $\Omega_2/2\pi$ (MHz) & $(c^\infty_1)^2$ & $(c^\infty_2)^2$ & $(c^\infty_3)^2$ \\
\hline\hline
0 & 200 & 1.0 & 0 & 0 \\
50 & 200 & 0.880 & 0.115 & $\leq 0.05$ \\
100 & 200 & 0.6134 & 0.3354 & 0.0511 \\
150 & 200 & 0.3660 & 0.4692 & 0.1646 \\
200 & 200 & 0.2069 & 0.4857 & 0.3073 \\
%\end{ruledtabular}
\hline \hline
\end{tabular}
|
58
|
2508.00607
| null |
\caption{Asymptotic dressed energies of the entrance and open exit channels with their main component and weight in the dressed adiabatic state as well as the branching ratio giving the percentage of a wavefunction to end up in a given exit channel.}
|
\label{S_matrix_table}
|
\begin{tabular}{cccc}
\hline
\hline
Energy& Main component & \multirow{2}{*}{Weight} & Branching\\ [0.2ex]
(MHz) & $\ket{j_1 |m_1|, j_2 |m_2|} \ket{N_1,N_2}$& & ratio\\[0.25ex]
\hline\hline
-0.0349 &$\ket{00,00} \ket{0,0}$ & 0.880 & -\\[0.25ex]
\hline
-0.0968 &$\ket{22,22} \ket{-2,2}$ &0.999 & 0\%\\[0.25ex]
\hline
-0.2320 &$\ket{00,22} \ket{-1,1}$ &0.922 & 0.006 \% \\[0.25ex]
\hline
-2.125 &$\ket{22,21} \ket{-2,2}$ & 0.815 & 0\%\\[0.25ex]
\hline
-2.257 &$\ket{00,21} \ket{-1,1}$ & 0.922&0.56\%\\[0.25ex]
\hline
-3.017 & $\ket{22,20} \ket{-2,2}$ & 0.922 & 0.16\% \\[0.25ex]
\hline
-3.100 & $\ket{00,20} \ket{-1,1}$ & 0.742 & 88.32\%\\[0.25ex]
\hline
-4.154 & $\ket{21,21} \ket{-2,2}$ & 0.735 & 0.37\% \\[0.25ex]
\hline
-5.046 & $\ket{21,20} \ket{-2,2}$&0.922 & 0.50 \%\\[0.25ex]
\hline
-5.9359 & $\ket{20,20} \ket{-2,2}$& 0.8532 & 10.06\% \\[0.25ex]
\hline \hline
\end{tabular}
|
59
|
2508.00607
| null |
\caption{Basis set of different $(|m_1|,|m_2|)$ sub-spaces for a $\pi-\pi$ polarized two-photon transition. For each subspace we give the involved values of the individual projections, followed by the symmetrized wavefunction and the Fock state. For each subspace we define a coupling scheme illustrated in the figures quoted in the last column.}
|
\label{Five_level_system_table}
|
\begin{tabular}{ccccc}
\hline
\hline
($|m_1|,|m_2|$)& $\ket{e_1,j_1,|m_1|,p_1,e_2,j_2,|m_2|,p_2}_+$ & $\ket{N_1, N_2}$& Coupling scheme& Figure\\ [0.5ex]
\hline%\hline
\multirow{5}{*}{($0,0$)} & $\ket{X,0,\mathbf{0},+1,X,0,\mathbf{0},+1}$& $\ket{0, 0}$& \multirow{5}{*}{Five level system}& \multirow{5}{*}{\ref{full_level_scheme}.a}\\
& $\ket{X,0,\mathbf{0},+1,X,2,\mathbf{0},+1}$& $\ket{-1, +1}$& & \\
& $\ket{X,2,\mathbf{0},+1,X,2,\mathbf{0},+1}$& $\ket{-2, +2}$& & \\[0.5ex]
& $\ket{X,0,\mathbf{0},+1,,b,1,\mathbf{0},-1}$& $\ket{-1, 0}$& & \\
& $\ket{X,2,\mathbf{0},+1,b,1,\mathbf{0},-1}$& $\ket{-2, +1}$& &\\ [0.25ex]
\hline
% \hline
\multirow{5}{*}{($0,1$)} & $\ket{X,0,\mathbf{0},+1,X,2,\mathbf{1},+1}$& $\ket{-1, +1}$&\multirow{5}{*}{Four level system}&\multirow{5}{*}{\ref{full_level_scheme}.b}\\
& $\ket{X,2,\mathbf{1},+1,X,2,\mathbf{0},+1}$& $\ket{-2, +2}$& & \\[0.5ex]
& $\ket{X,0,\mathbf{0},+1,b,1,\mathbf{1},-1}$& $\ket{-1, 0}$& & \\
& $\ket{X,2,\mathbf{1},+1,b,1,\mathbf{0},-1}$& $\ket{-2, +1}$& & \\
& $\ket{X,2,\mathbf{0},+1,b,1,\mathbf{1},-1}$& $\ket{-2, +1}$& & \\[0.25ex]
\hline
%\hline
\multirow{3}{*}{($0,2$)} & $\ket{X,0,\mathbf{0},+1,X,2,\mathbf{2},+1}$& $\ket{-1, +1}$& \multirow{3}{*}{Three level $\Lambda$ system}&\multirow{3}{*}{\ref{full_level_scheme}.c}\\
& $\ket{X,2,\mathbf{2},+1,X,2,\mathbf{0},+1}$& $\ket{-2, +2}$& & \\[0.5ex]
& $\ket{X,2,\mathbf{2},+1,b,1,\mathbf{0},-1}$& $\ket{-2, +1}$& & \\
\hline
%\hline
\multirow{2}{*}{($1,1$)} & $\ket{X,2,\mathbf{1},+1,X,2,\mathbf{1},+1}$& $\ket{-2, +2}$& \multirow{2}{*}{Single field coupling}& \multirow{2}{*}{\ref{full_level_scheme}.d} \\[0.5ex]
& $\ket{X,2,\mathbf{1},+1,b,1,\mathbf{1},-1}$& $\ket{-2, +1}$& & \\[0.25ex]
\hline
%\hline
\multirow{2}{*}{($1,2$)} & $\ket{X,2,\mathbf{1},+1,X,2,\mathbf{2},+1}$& $\ket{-2, +2}$&\multirow{2}{*}{Single field coupling}&\multirow{2}{*}{\ref{full_level_scheme}.e}\\
& $\ket{X,2,\mathbf{1},+1,b,1,\mathbf{1},-1}$& $\ket{-2, +1}$& & \\[0.25ex]
\hline
%\hline
($2,2$)& $\ket{X,2,\mathbf{2},+1,X,2,\mathbf{2},+1}$& $\ket{-2, +2}$&No coupling& \ref{full_level_scheme}.f\\
\hline
\hline
\end{tabular}
|
60
|
2508.05616v1
| null |
\caption{Comparison of \our{} with trajectory prediction heuristics across datasets with mean minADE$_{20}$ / minFDE$_{20}$ (meters) on the ETH-UCY dataset.}
|
\label{table:main_hs}
|
\begin{tabular}{l||c|c|c|c|c||c}
\toprule
Method & ETH & HOTEL & UNIV & ZARA1 & ZARA2 & AVG \\
\midrule
SocialForce \cite{helbing1995social} & 1.46/2.48 & 0.69/1.23 & 0.96/1.75 & 1.37/2.51 & 0.84/1.53 & 1.06/1.90 \\
LinReg \citep{bishop2006pattern} & 1.04/2.20 & 0.26/0.47 & 0.76/1.48 & 0.62/1.22 & 0.47/0.93 & 0.63/1.26 \\
ConstantAcc \citep{polychronopoulos2007sensor} & 3.12/7.98 & 1.64/4.19 & 1.02/2.60 & 0.81/2.05 & 0.60/1.53 & 1.44/3.67 \\
CSCRCTR \cite{s140305239} & 2.27/4.61 & 1.03/2.18 & 1.35/3.12& 0.96/2.12 & 0.90/2.10 & 1.30/2.83 \\
CVM \cite{scholler2020constant} & 1.01/2.24 & 0.32/0.61 & 0.54/1.21 & 0.42/0.95 & 0.33/0.75 & 0.52/1.15 \\
CVM-S \cite{scholler2020constant} & 0.92/2.01 & 0.27/0.51 & 0.53/1.17 & 0.37/0.77 & 0.28/0.63 & 0.47/1.02 \\
CTRV \cite{lu2021ctrv} & 1.62/3.64 & 0.72/1.09 & 0.71/1.59 & 0.65/1.50 & 0.48/1.10 & 0.84/1.78 \\
\midrule
\our{}
& \textbf{0.47/0.77}
& \textbf{0.17/0.30}
& \textbf{0.51/1.10}
& \textbf{0.35/0.75}
& \textbf{0.27/0.57}
& \textbf{0.35/0.70} \\
\bottomrule
\end{tabular}
|
61
|
2508.05616v1
| null |
\caption{Comparison of \our{} with deep learning approaches (mean minADE$_{20}$/minFDE$_{20}$ on ETH-UCY). Each \underline{underlined} number indicates that the result is worse than the corresponding result from \our{} on at least one metric.}
|
\label{table:main_nn}
|
\begin{tabular}[\linewidth]{l||c|c|c|c|c||c}
\toprule
Method & ETH & HOTEL & UNIV & ZARA1 & ZARA2 & AVG \\
\midrule
Social-LSTM \cite{alahi2016social}
& \underline{1.09/2.35}
& \underline{0.79/1.76}
& \underline{0.67/1.40}
& \underline{0.56/1.17}
& \underline{0.72/1.54}
& \underline{0.77/1.64} \\
Social-GAN \cite{gupta2018social}
& \underline{0.87/1.62}
& \underline{0.67/1.37}
& \underline{0.76/1.52}
& 0.35/0.68
& \underline{0.42/0.84}
& \underline{0.61/1.21} \\
STGAT \cite{Huang_2019_ICCV}
& \underline{0.65/1.12}
& \underline{0.35/0.66}
& \underline{0.52/1.10}
& 0.34/0.69
& \underline{0.29/0.60}
& \underline{0.43/0.83} \\
Social-STGCNN \cite{mohamed2020social}
& \underline{0.64/1.11}
& \underline{0.49/0.85}
& 0.44/0.79
& 0.34/0.53
& \underline{0.30/0.48}
& \underline{0.44/0.75} \\
Trajectron++ \cite{salzmann2020trajectron++}
& \underline{0.61/1.03}
& \underline{0.20}/0.28
& 0.30/0.55
& 0.24/0.41
& 0.18/0.32
& 0.31/0.52 \\
MemoNet \cite{xu2022remember}
& 0.41/0.61
& \textbf{0.11}/\textbf{0.17}
& 0.24/0.43
& 0.18/0.32
& 0.14/0.24
& 0.21/0.35 \\
EigenTrajectory \cite{bae2023eigentrajectory}
& \textbf{0.36}/\textbf{0.53}
& 0.12/0.19
& 0.24/0.43
& 0.19/0.33
& 0.14/0.24
& 0.21/0.34 \\
MoFlow \cite{fu2025moflow}
& 0.40/0.57
& \textbf{0.11}/\textbf{0.17}
& \textbf{0.23}/\textbf{0.39}
& \textbf{0.15}/\textbf{0.26}
& \textbf{0.12}/\textbf{0.22}
& \textbf{0.20}/\textbf{0.32} \\
\midrule
\our{}
& 0.47/0.77
& 0.17/0.30
& 0.51/1.10
& 0.35/0.75
& 0.27/0.57
& 0.35/0.70 \\
\bottomrule
\end{tabular}
|
73
|
2508.00466
| null |
\caption{Selection of rare and exclusive SM Higgs decays potentially observable in pp(14~TeV) collisions at the HL-LHC (or for which a limit could be ``easily'' set from existing data).
For each decay, we list the theoretical branching fraction(s), the current experimental limit and the conservative bounds estimated for HL-LHC (as well as the theory over HL-LHC-expected ratio, $\BR(\rm th)/\BR(\text{exp,HL-LHC})$).
%The last column indicates the approximate ratio of theoretical over our conservative extrapolated rates, $\BR(\rm th)/\BR(\text{exp,HL-LHC})$.
\label{tab:hl_lhc_summary}}
| null | null |
17,127
|
2501.18324
| null |
\caption{Dialogues created by three separate annotator pairs in VDAct for scenario IDs 00265 and 00523.}
|
\label{tab:example-dials}
|
\begin{tabular}{c|p{0.28\linewidth}|p{0.28\linewidth}|p{0.28\linewidth}}
\hline \hline
\multicolumn{4}{l}{\textbf{Example}\#1, \textbf{Scenario ID}: 00265, \textbf{Activities}: Pick up dirty fork1, Pick up dirty waterglass1, Clean sink4, Put groceries in fridge23, Fall while preparing meal1} \\ \hline
& \textbf{Dialogue ID}: 0026501 & \textbf{Dialogue ID}: 0026502 & \textbf{Dialogue ID}: 0026503 \\ \hline
\multirow{2}{*}{1} & Was the man carrying anything when he fell? & What room did the man start his activities in? & Where does he begin at today? \\ \cline{2-4}
& Yes, he was. & In the kitchen. & He finds himself in the kitchen manhandling some forks. \\ \hline
\multirow{2}{*}{2} & Was he carrying a frying pan? & What is the first thing he did there? & Why is he handling forks? \\ \cline{2-4}
& Yes, he had a pan in his right hand. & He picks up two dirty forks. & I guess they are dirty and decides to take them to the sink. \\ \hline
\multirow{2}{*}{3} & Was he moving to or from the stove when he fell? & What does he do with them? & Does he use water to wash them or just sets them there? \\ \cline{2-4}
& He was facing the direction of the stove, but it's unclear where he was taking the items he was holding. & He places them in the sink. & He just sets them there and then goes back to the table for some glasses. \\ \hline
\multirow{2}{*}{4} & Was he cooking something on the stove? & What does he do next? & What does he do with the glasses? \\ \cline{2-4}
& No, nothing is sitting on the burners. & He places dirty water glasses in the sink too. & He also takes them to the sink and maybe rinses them out but I can't see any running water. \\ \hline
\multirow{2}{*}{5} & How many items does the man put into the fridge? & Does he do anything else when he's at the sink? & Okay, so after this what does he decide to get into? \\ \cline{2-4}
& One item. & He also cleans the sink. & Well the glasses and forks vanish and decides to sponge down the sink. \\ \hline
\multirow{2}{*}{6} & Did the man put a non-food item into the fridge? & What does he use to clean it? & Does he get the sponge wet first? \\ \cline{2-4}
& No, it is an edible item. & A sponge. & No he dry wipes first and then uses some water and wipes again. \\ \hline
\multirow{2}{*}{7} & Where does he put the dirty water glass? & Does he do anything else in the kitchen after cleaning the sink? & How long did it take him to do this? \\ \cline{2-4}
& In the kitchen sink. & He cleans a carton of milk before putting it in the fridge. & Just a second as he got distracted by the cartoon of TruMoo sitting out behind the sink. \\ \hline
\multirow{2}{*}{8} & Did he fill it with water after putting it in the sink? & What else does he put in the fridge? & What's he do with the TruMoo? \\ \cline{2-4}
& Nope. & He doesn't put anything else in the fridge. & He goes to get a towel to wipe it down and then he puts in the sink. \\ \hline
\multirow{2}{*}{9} & How many dirty forks does the man pick up? & What does he do next? & So after he finishes that task where does he go? \\ \cline{2-4}
& Two dirty forks. & He picks up a frying pan and package of chicken before falling. & He decides he needs to cook something so he grabs a pan. \\ \hline
\multirow{2}{*}{10} & Did he get them from the living room? & How does he fall? & What does he cook in the pan? \\ \cline{2-4}
& No, they were in the kitchen. & He falls forward onto his face. & He never makes it that far as he face plants and then lays on the floor forever. \\ \hline \hline
\multicolumn{4}{l}{} \\
\multicolumn{4}{l}{\textbf{Example}\#2, \textbf{Scenario ID}: 00523, \textbf{Activities}: Drink alcohol1, Pick up dirty dishes1, Relax on sofa1, Fall while standing quietly1, Cook salmon1} \\ \hline
& \textbf{Dialogue ID}: 0052301 & \textbf{Dialogue ID}: 0052302 & \textbf{Dialogue ID}: 0052303 \\ \hline
\multirow{2}{*}{1} & What room do you see the man in initially? & What activity did the man start by doing? & What is his first action? \\ \cline{2-4}
& Initially, the living room, but it looks like he was moving through the living room to the kitchen. & He drank some alcohol first. & Today he is in the kitchen looking for a drink. \\ \hline
\multirow{2}{*}{2} & Does he make it to the kitchen safely? & What kind of alcohol did he drink? & Does he find one? \\ \cline{2-4}
& Yes, where liquid refreshment awaits him. & Something from a clear bottle of liqueur. & He does yes, his favorite curacao, he takes a sip. \\ \hline
\multirow{2}{*}{3} & What kind of beverage is there for him to enjoy? & Where did he drink it? & What does he do after drinking? \\ \cline{2-4}
& Some kind of alcohol in a clear bottle. & In the kitchen. & He checks out the plates on the table, they seem dirty. \\ \hline
\multirow{2}{*}{4} & How many total sips does he have of the alcohol? & Was his next activity in the kitchen too? & How does he clean them then? \\ \cline{2-4}
& Two sips plus a long pull on the bottle. & Yes, it was. & Well he takes them to the sink for a nice soaking. \\ \hline
\multirow{2}{*}{5} & What is next on the man's agenda? & What did he do in the kitchen? & How long does he soak them for? \\ \cline{2-4}
& He relocates some dirty dishes to the sink. & He picked up two dirty dishes and put them in the sink. & I guess the rest of the day because he's off to the living room now for a rest. \\ \hline
\multirow{2}{*}{6} & Does he remain in the kitchen for any other reason? & What did the man do after that? & Where does he rest there? \\ \cline{2-4}
& No, he bee lines for the living room after depositing the plates in the sink. & Next, he went into the living room and relaxed on the sofa for a bit. & On the couch. \\ \hline
\multirow{2}{*}{7} & What awaits him in the living room? & What else did he do in the living room? & What does he do after resting? \\\cline{2-4}
& Relaxation on his sofa. & He grabbed a couple of books off his bookshelf but had a bit of an issue after that. & He gets up and heads over to the book shelf and grabs two books. \\ \hline
\multirow{2}{*}{8} & Does he stay there for the remainder of his day? & What was his issue? & What were the books could you tell? \\ \cline{2-4}
& No, but he stays for a while longer because he decided to read a couple of books. & He fell over while holding the books. & I couldn't tell because he falls flat on his face as soon as he gets the book from the shelf. \\ \hline
\multirow{4}{*}{9} & Does he sit to read those chosen books? & Did he recover and move on to his last activity? & Is he able to get back up though? \\ \cline{2-4}
& Don't know because he just all of a sudden falls over while standing there holding the books. & Yes, he went into the kitchen to do some cooking. & Yes, and he quickly heads to the kitchen for some cooking now. \\ \hline
\multirow{2}{*}{10} & Does he recover from his fall? & What was he cooking in the kitchen? & What does he decide to cook? \\ \cline{2-4}
& Yes, and heads into the kitchen to do some cooking. & A packet of salmon. & He's got his eye on some salmon. He washes it off in the package and takes it to the stove. \\ \hline
\multirow{2}{*}{11} & What does he cook? & How did he cook it? & -- \\ \cline{2-4}
& He is cooking some salmon. & On the stove, using a frying pan. & -- \\ \hline
\end{tabular}
|
62
|
2508.05616v1
| null |
\caption{Out of distribution performance of methods trained on different ETH-UCY splits and tested on the unseen SDD dataset. We report minADE$_{20}$ / minFDE$_{20}$ (pixels) on the SDD dataset.}
|
\label{tab:cross-dataset-generalization}
|
\begin{tabular}{l c c c c c c}
\toprule
Method (tested on SDD dataset) & ETH & HOTEL & UNIV & ZARA1 & ZARA2 & AVG \\
\midrule
SocialForce \cite{helbing1995social}
& 33.64/60.63 & 33.64/60.63 & 33.64/60.63 & 33.64/60.63 & 33.64/60.63 & 33.64/60.63 \\
CVM \cite{scholler2020constant}
& 18.82/37.95 & 18.82/37.95 & 18.82/37.95 & 18.82/37.95 & 18.82/37.95 & 18.82/37.95\\
CVM-S \cite{scholler2020constant}
& 16.28/31.84 & 16.28/31.84 & 16.28/31.84 & 16.28/31.84 & 16.28/31.84 & 16.28/31.84 \\
\midrule
Trajectron++ \cite{salzmann2020trajectron++}
& 46.72/69.11 & 47.30/67.76 & 46.08/75.90 & 47.30/72.19 & 46.78/68.59 & 46.84/70.71 \\
EigenTrajectory \cite{bae2023eigentrajectory}
& 14.51/25.13 & 14.69/24.64 & 14.31/27.60 & 14.69/26.25 & 14.53/24.94 & 14.55/25.71 \\
MoFlow \cite{fu2025moflow}
& 17.00/27.98 & 17.21/27.43 & 17.00/30.63 & 17.24/29.22 & 17.27/27.56 & 17.14/28.56 \\
\midrule
\our{}
& \textbf{12.58}/\textbf{23.82} & \textbf{12.26}/\textbf{23.60} & \textbf{12.78}/\textbf{23.71} & \textbf{13.10}/\textbf{25.23} & \textbf{12.22}/\textbf{23.57} & \textbf{12.59}/\textbf{23.99} \\
\bottomrule
\end{tabular}
|
63
|
2508.05616v1
| null |
\caption{Ablation study for the evolution framework removing different components. Lower is better $(\downarrow)$.}
|
\label{tab:ablation}
|
\begin{tabular}{@{}lccc@{}}
\toprule
\setlength{\tabcolsep}{1mm}
{Dataset} & {~ - SFL - CGES} & {~ - SFL} & {\our{}} \\
\midrule
ETH & 0.68/1.36 & 0.59/1.12 & \textbf{0.47/0.77} \\
HOTEL & 0.26/0.45 & 0.19/0.33 & \textbf{0.17/0.30} \\
UNIV & 0.59/1.22 & 0.52/1.13 & \textbf{0.51/1.10} \\
ZARA1 & 0.37/0.77 & 0.36/0.76 & \textbf{0.35/0.75} \\
ZARA2 & 0.31/0.65 & 0.28/0.59 & \textbf{0.27/0.57} \\
\bottomrule
\end{tabular}
|
64
|
2508.05616v1
| null |
\caption{Performance of various LLMs on the standard ETH-UCY benchmark datasets. Best results are in bold.}
|
\label{tab:llm}
|
\begin{tabular}{l cc cc cc}
\toprule
% First row of datasets
\textbf{Model} & \multicolumn{2}{c}{\textbf{ETH}} & \multicolumn{2}{c}{\textbf{HOTEL}} & \multicolumn{2}{c}{\textbf{UNIV}} \\
\cmidrule(lr){2-3} \cmidrule(lr){4-5} \cmidrule(lr){6-7}
& ADE & FDE & ADE & FDE & ADE & FDE \\
\midrule
deepseek-R1 & 0.46 & 0.74 & 0.16 & \textbf{0.29} & \textbf{0.49} & \textbf{1.06} \\
deepseek-V3 & \textbf{0.43} & \textbf{0.71} & 0.21 & 0.39 & 0.50 & 1.08 \\
Qwen3-32B & 0.49 & 0.74 & 0.18 & 0.32 & 0.51 & 1.09 \\
ChatGPT-4o & 0.51 & 0.82 & \textbf{0.15} & 0.30 & 0.54 & 1.13 \\
Claude-3.7 & 0.49 & 0.93 & 0.18 & 0.36 & \textbf{0.49} & 1.07 \\
\midrule[1pt] % Separator line
% Second row of datasets
\textbf{Model} & \multicolumn{2}{c}{\textbf{ZARA1}} & \multicolumn{2}{c}{\textbf{ZARA2}} & \multicolumn{2}{c}{\textbf{AVG}} \\
\cmidrule(lr){2-3} \cmidrule(lr){4-5} \cmidrule(lr){6-7}
& ADE & FDE & ADE & FDE & ADE & FDE \\
\midrule
deepseek-R1 & 0.33 & 0.72 & \textbf{0.26} & \textbf{0.54} & \textbf{0.34} & \textbf{0.67} \\
deepseek-V3 & 0.34 & \textbf{0.69} & 0.27 & 0.56 & 0.35 & 0.69 \\
Qwen3-32B & 0.35 & 0.75 & 0.27 & 0.59 & 0.36 & 0.70 \\
GPT-4o & 0.36 & 0.72 & 0.31 & \textbf{0.54} & 0.37 & 0.70 \\
Claude-3.7 & \textbf{0.32} & \textbf{0.61} & 0.27 & 0.57 & 0.35 & 0.71 \\
\bottomrule
\end{tabular}
|
65
|
2508.05616v1
| null |
\caption{Main hyperparameters for the TRAJEVO framework.}
|
\label{tab:hyperparams}
|
\begin{tabular}{@{} p{5.5cm} l @{}}
\toprule
\textbf{Hyperparameter} & \textbf{Value} \\
\midrule
\multicolumn{2}{@{}l}{\textbf{Evolutionary Algorithm}} \\
\quad Population size & 10 \\
\quad Number of initial generation & 8 \\
\quad Elite ratio for crossover & 0.3 \\
\quad Crossover rate & 1 \\
\quad Mutation rate & 0.5 \\
\quad CGES Softmax temperature & 1.0 \\
\addlinespace
\multicolumn{2}{@{}l}{\textbf{Large Language Model (LLM)}} \\
\quad LLM model & Gemini 2.0 Flash \\
\quad LLM temperature (generator \& reflector) & 1 \\
\quad Max words for short-term reflection & 200 words \\
\quad Max words for long-term reflection & 20 words \\
\addlinespace
\multicolumn{2}{@{}l}{\textbf{Trajectory Prediction}} \\
\quad Num. prediction samples ($K$) & 20 \\
\quad Observation length ($T_{\text{obs}}$) & 8 frames (3.2s) \\
\quad Prediction length ($T_{\text{pred}}$) & 12 frames (4.8s) \\
\bottomrule
\end{tabular}
|
66
|
2508.00466
| null |
\caption{Rare Higgs decays to two neutrinos, a photon plus two neutrinos, three or four photons, and a Z boson plus two $\gamma$'s or two gluons. For each decay, we provide the theoretical branching fraction computed with \mgshort. No current (or extrapolated, future) experimental limits exist.
\label{tab:H_decays_V_V_V}}
| null | null |
67
|
2508.00466
| null |
\caption{Exclusive Higgs decay rates into a photon plus a vector meson. For each decay, we provide the theoretical branching fraction(s), the ratios of direct-over-indirect and of exclusive-direct-over-inclusive decay amplitudes, the current experimental limit and that conservatively estimated for HL-LHC (as well as the theory over HL-LHC-expected ratio). %The last two columns indicate the ratios of direct-over-indirect and of exclusive-direct-over-inclusive amplitudes.
\label{tab:H_decays_gamma_meson}}
| null | null |
68
|
2508.00466
| null |
\caption{Exclusive Higgs decay rates into a Z boson plus a meson. For each decay, we provide the theoretical branching fraction(s), the relative size of the direct contribution to the total branching fraction ($\delta_\text{dir}$), the exclusive-direct-to-inclusive decay amplitude ratio, the current experimental limit and that conservatively estimated for HL-LHC (as well as the theory over HL-LHC-expected ratio). %The last two columns indicate the ratios of direct-over-indirect and of exclusive-direct-over-inclusive amplitudes.
\label{tab:H_decays_Z_meson}}
| null | null |
69
|
2508.00466
| null |
\caption{Exclusive Higgs decay rates into a neutral gauge boson plus a flavoured meson. For each decay, we provide the branching fraction derived via Eqs.~(\ref{eq:BR_H_gamma_m})--(\ref{eq:BR_H_Z_m}), as well as the current experimental limit and that conservatively estimated for HL-LHC (as well as the SM over HL-LHC-expected ratio). %The last two columns indicate the ratios of direct-over-indirect and of exclusive-direct-over-inclusive amplitudes.
\label{tab:H_decays_boson_flavouredmeson}}
| null | null |
70
|
2508.00466
| null |
\caption{Exclusive Higgs decay rates into a W boson plus a meson. For each decay, we provide the theoretical branching fraction(s), and the relative size of the direct contribution to the total branching fraction ($\delta_\text{dir}$). No current experimental limits, nor future estimates for them, exist.
\label{tab:H_decays_W_meson}}
| null | null |
71
|
2508.00466
| null |
\caption{Exclusive Higgs decay rates to a photon or a Z boson plus an ortho- $(\lele)_1$ or para- $(\lele)_0$ leptonium state (only the ground states, $n=1$, are considered). For each decay, we provide the theoretical SM branching fraction predictions. No current experimental limits, nor future estimates for them, exist.
%, as well as the current experimental limit and that conservatively estimated for HL-LHC (as well as the theory over HL-LHC-expected ratio).
\label{tab:H_decays_boson_leptonium}}
| null | null |
72
|
2508.00466
| null |
\caption{Exclusive Higgs decay rates to a pair of mesons. For each decay, we list the theoretical branching fraction(s), the current experimental limit and the conservative bounds estimated for HL-LHC (as well as the theory over HL-LHC-expected ratio).
\label{tab:H_2meson_decays}}
| null | null |
74
|
2508.00667
| null |
\caption{$R_\mathrm{200b}$ used for scaling when determining the radial range for fitting an NFW profile to the measured ESD profile. These radii are the mean $R_\mathrm{200b}$ in each $L_\mathrm{c}$ and $L_\mathrm{gap}$ bin extracted from the Illustris TNG300 simulation at $z=0$.}
|
\label{tab:R200b}
|
\begin{tabular}{cccc}
\hline\hline
\multicolumn{4}{c}{$R_\mathrm{200b}[h^{-1}\mathrm{Mpc}]$}\\\hline
&
$9.9<\textrm{log}L_\mathrm{c}[{h}^{-2}L_{\odot}]\leq 10.3$&$10.3<\textrm{log}L_\mathrm{c}[{h}^{-2}L_{\odot}]\leq 10.7$&$10.7<\textrm{log}L_\mathrm{c}[{h}^{-2}L_{\odot}]\leq 11.1$\\
$0<\textrm{log}L_\mathrm{gap}\leq 0.3$& $0.399$&$0.620$&$0.953$\\
$0.3<\textrm{log}L_\mathrm{gap}\leq 0.6$& $0.369$&$0.580$&$0.922$\\
$0.6<\textrm{log}L_\mathrm{gap}\leq 1$& $0.338$& $0.537$&...\\
\hline
\end{tabular}
|
75
|
2508.00667
| null |
\caption{The significances of models with or without the gap dependence in the halo parameters. The $p$-values indicate the probability that fitting with the assumed model is consistent with fitting statistical fluctuations in the null model. See details in Section~\ref{sec:significance}.}
|
\label{tab:significance}
|
\begin{tabular}{ccc}
\hline\hline
Models & Gap dependences&p-value\\\hline
M1&
$M(L_\mathrm{c}, L_{\rm gap})$, $c(L_\mathrm{c}, L_{\rm gap})$&$1.8\times 10^{-6}$\\
M2& $M(L_\mathrm{c}, L_{\rm gap})$, $c(L_\mathrm{c})$&$1.3\times 10^{-4}$\\
M3& $M(L_\mathrm{c})$, $c(L_\mathrm{c}, L_{\rm gap})$&$4.6\times 10^{-1}$\\
M4 (null model)& $M(L_\mathrm{c})$, $c(L_\mathrm{c})$& ...\\
\hline
\end{tabular}
|
76
|
2508.00658
| null |
\caption{Performance Comparison Across Dataset Types (Accuracy and F1-Scores). The elements in the table are accuracy in every row ecept the last row that the elements are F1-scores.}
|
\label{tab:performance_comparison}
|
\begin{tabular}{|p{2.0cm}|c|c|c|c|c|c|c|}
\hline
\multicolumn{1}{|c|}{} & \multicolumn{7}{c|}{\textbf{Accuracies of Methods}} \\
\hline
\textbf{Datasets} & \textbf{MB-VL} & \textbf{VLGC} & \textbf{GC} & \textbf{TE} & \textbf{VLTE} & \textbf{PCMCI+} & \textbf{GG} \\
\hline
Following relation & 0.833 & 0.733 & \textbf{1.000} & 0.367 & 0.933 & \textbf{1.000} & 0.400 \\
\hline
Variable-lag & 0.767 & 0.600 & 0.233 & 0.333 & 0.867 & \textbf{1.000} & 0.467 \\
\hline
Broadband lag & 0.867 & 0.433 & 0.833 & 0.467 & 0.933 & \textbf{0.967} & 0.000 \\
\hline
Multifrequency lag & \textbf{0.933} & 0.167 & 0.267 & 0.567 & 0.533 & 0.900 & 0.133 \\
\hline
Random Noise & 0.750 & \textbf{1.000} & \textbf{1.000} & 0.592 & 0.617 & 0.300 & 0.525 \\
\hline
\hline
\textbf{Overall F1-score} & \textbf{0.810} & 0.742 & 0.792 & 0.512 & 0.717 & 0.725 & 0.388 \\
\hline
\end{tabular}
|
77
|
2508.00658
| null |
\caption{Performance Impact of Band Configuration Strategy}
|
\label{tab:band_strategy}
|
\begin{tabular}{|l|c|c|c|}
\hline
\textbf{Configuration} & \textbf{Strategy} & \textbf{Multi-freq} & \textbf{Overall F1} \\
\hline
Single Band & Broadband analysis & 0.167 & 0.742\\
\hline
Two Bands & Optimal balance & 0.933 & 0.810 \\
\hline
EEG Bands & Frequency-specific & 1.000 & 0.618 \\
\hline
\end{tabular}
|
78
|
2508.00658
| null |
\caption{Lag Detection Performance by Frequency Band}
|
\label{tab:lag_performance}
|
\begin{tabular}{|l|c|c|c|c|}
\hline
\textbf{Band} & \textbf{True Lag} & \textbf{Inferred Lag} & \textbf{Sig. Lag} & \textbf{Error} \\
\hline
Alpha & 15 & $12.3 \pm 6.0$ & $11.6 \pm 7.0$ & $1.2 \pm 0.4$ \\
\hline
L Gamma & 8 & $4.3 \pm 3.9$ & $5.0 \pm 0.0$ & $3.0 \pm 0.0$ \\
\hline
H Gamma & 4 & $2.4 \pm 3.3$ & $2.4 \pm 3.3$ & $2.5 \pm 1.8$ \\
\hline
\end{tabular}
|
79
|
2508.00658
| null |
\caption{Causality Detection Results on Real-world Datasets}
|
\label{tab:realworld_results_minimal}
|
\begin{tabular}{|l|c|c|c|c|c|c|c|}
\hline
& \multicolumn{7}{c|}{\textbf{Methods}} \\
\hline
\textbf{Case} & \textbf{MB-VL} & \textbf{VLGC} & \textbf{G} & \textbf{TE} & \textbf{VLTE} & \textbf{PCMCI} & \textbf{GG} \\
\hline
EEG & 1 & 0 & 0 & 1 & 1& 1 & 0\\
\hline
Chick. Egg & 1 & 1 & 1 & 1 & 1 & 1 & 1\\
\hline
Old ffg. & 1 & 1 & 0 & 0 & 1 & 1 & 0 \\
\hline
Gas fur. & 1 & 1 & 1 & 1 & 1 & 1 & 1\\
\hline
\end{tabular}
|
80
|
2508.00523
| null |
\caption{Summary of results for online nonsubmodular optimization under different settings, where $n$ is the dimensionality, $d$ is the maximum delay and $\Bar{d} = \frac{1}{T}\sum_{t=1}^T d_t$ is the average delay over $T$ rounds. For simplicity, we use the abbreviations: full $\rightarrow$ full-information setting with delayed feedback, bandit $\rightarrow$ bandit setting with delayed feedback.}
|
\label{tab:1}
|
\begin{tabular}{cccc}
\toprule
Setting & Method & $(\alpha,\beta)$-regret bound \\
\midrule
bandit & DBAGD \cite{lin2022online}& $\mathcal{O}\left(nd^{1/3}T^{2/3}\right)$ \\
\midrule
bandit & BDGD-NF (\textbf{Theorem} \ref{thm:2})& $\mathcal{O}\left(n\Bar{d}^{1/3}T^{2/3}\right)$ \\
\midrule
bandit & BDBGD-NF (\textbf{Theorem} \ref{thm:1}) & $\mathcal{O}\left(n(T^{2/3}+\sqrt{dT})\right)$ \\
\midrule
full & DOAGD \cite{lin2022online}& $\mathcal{O}\left(\sqrt{ndT}\right)$ \\
\midrule
full & DOGD-NF (\textbf{Theorem} \ref{thm:0}) & $\mathcal{O}\left(\sqrt{n\Bar{d}T}\right)$ \\
\bottomrule
\end{tabular}
|
81
|
2508.00701
| null |
\caption{
Detection results on GenVideo datasets.
Our D3 is training-free, while the baselines are trained on real videos from Youku-mPLUG~\cite{xu2023youku} and AI-generated videos from Pika~\cite{pikaPika}, following the setting in Demamba~\cite{chen2024demamba}. \textbf{Bold} represents the best and \underline{underline} represents the second best.
}
|
\label{tab:genvideo}
|
\begin{tabular}{lccccccccccccc}
\toprule
Detection& Detection & \multicolumn{10}{c}{Datasets (AP$\uparrow$)} & \multirow{2}{*}{mAP}\\
\cmidrule(r){3-12}
Method & Level & Crafter& Gen2 & HotShot & Lavie & MSE & MV & MSO & Show-1 & Sora & WS \\
\midrule
FID~\cite{zheng2025breaking} & Image & 92.41 & 93.27 & \underline{86.10} & 83.68 & \underline{91.50} & 93.67 & 92.24 & \underline{90.61} & 74.95 & \underline{82.24} & \underline{88.07}\\
NPR~\cite{tan2023rethinking} & Image & 97.02 & 96.35 & 40.17 & 22.37 & 84.67 & 96.79 & 96.53 & 21.61 & \underline{90.55} & 66.51 & 71.26\\
STIL~\cite{gu2021spatiotemporal} & Image & 85.82 & 93.19 & 40.61 & 53.24 & 58.99 & 94.94 & 71.62 & 47.73 & 22.35 & 61.91 & 63.04 \\
MINITIME~\cite{coccomini2024mintime} & Video & 88.62 & 60.66 & 39.03 & 82.29 & 23.85 & 74.79 & 74.33 & 41.08 & 16.92 & 72.25 & 57.38 \\
FTCN~\cite{zheng2021exploring} & Video & 95.41 & 97.18 & 37.47 & 44.90 & 79.71 & 99.75 & 97.05 & 17.33 & 83.69 & 66.86 & 71.94\\
TALL~\cite{xu2023tall} & Video & 87.85 & 93.47 & 44.00 & 59.07 & 51.11 & 92.09 & 63.63 & 51.06 & 15.82 & 64.43 & 62.25\\
XCLIP~\cite{ni2022expanding} & Video & {97.32} & \textbf{99.44} & 44.68 & 72.69 & 88.00 & \textbf{99.96} & 97.53 & 38.37 & 71.08 & 74.00 & 78.31\\
AIGVDet~\cite{bai2024ai} & Video & 75.87 & 89.98 & {51.81} & \underline{88.62} & 70.91 & 56.22 & 67.93 & 72.59 & 65.70 & 64.96 & 70.46\\
Demamba~\cite{chen2024demamba} & Video & \underline{97.91} & {99.16} & {52.97} & 76.72 & 82.83 & \underline{99.80} & \underline{98.42} & 56.24 & 77.75 & 74.81 & 81.66\\
Our D3 & Video & \textbf{98.53} & \underline{99.39} & \textbf{98.52} & \textbf{97.22} & \textbf{97.12} & 99.52 & \textbf{98.68} & \textbf{99.18} & \textbf{99.91} & \textbf{96.49} & \textbf{98.46} \\
\bottomrule
\end{tabular}
|
95
|
2508.00380
| null |
\caption{The number of generations to complete each benchmark by each method. \textbf{Note:} sequential methods are not listed here since they use exactly the same number of generations as FEs.}
|
\label{tab:iters}
|
\begin{tabular}{l | c c c c c}
\hline
Benchmark & TuRBO & CMA-ES & PSO & Nelder-Mead & \MethodName \\
\hline
Numerical & 10$\sim$100 & 83$\sim$526 & 25$\sim$50 & 956$\sim$9395 & \textbf{10} \\
Landing & \textbf{10} & 91 & 25 & 823$\sim$901 & \textbf{10} \\
Pushing & \textbf{10} & 91 & 25 & 862$\sim$933 & \textbf{10} \\
Rover & 100 & 625 & 25 & 3163$\sim$3216 & \textbf{10} \\
Walker & 100 & 526 & 50$\sim$51 & 1833$\sim$1839 & \textbf{10} \\
Ant & 100 & 455 & 24$\sim$25 & 1500$\sim$1503 & \textbf{10} \\
\hline
\end{tabular}
|
82
|
2508.00701
| null |
\caption{
Detection results on 14 EvalCrafter datasets.
}
|
\label{tab:evalcrafter}
|
\begin{tabular}{lcccccccccccccccc}
\toprule
Detection & \multicolumn{14}{c}{Datasets (AP$\uparrow$)} & \multirow{2}{*}{mAP}\\
\cmidrule(r){2-15}
Method & MV & Floor32 & Gen2 & Gen2-D & HotShot & LaVie-V & LaVie-I & Mix-SR & MSE & Pika & Pika-v1 & Show-1 & VC & ZS \\
\midrule
FID & 98.29 & 96.4 & 97.36 & 98.68 & \underline{89.9} & \underline{92.92} & \underline{84.19} & 98.51 & 95.74 & \underline{99.49} & 99.17 & \underline{96.77} & 95.71 & 95.18 & \underline{95.59} \\
NPR & \textbf{99.96} & \textbf{99.77} & \underline{99.34} & \textbf{99.95} & 47.39 & 76.45 & 72.23 & \textbf{99.67} & \textbf{98.54} & \textbf{99.97} & \textbf{99.93} & 69.82 & \textbf{99.68} & \underline{98.21} & 90.07 \\
AIGVDet & 56.50 & 67.84 & 71.86 & 74.24 & 51.46 & 73.81 & 70.72 & 57.64 & 71.00 & 94.95 & 92.92 & 72.41 & 64.58 & 67.00 & 70.50 \\
Demamba & 99.49 & 91.76 & 96.98 & 99.27 & 34.60 & 56.89 & 37.85 & 97.49 & 71.33 & 98.69 & 99.33 & 26.83 & 94.30 & 64.39 & 76.37 \\
Our D3 & \underline{99.52} & \underline{98.68} & \textbf{99.46} & \underline{99.74} & \textbf{98.52} & \textbf{97.79} & \textbf{98.48} & \underline{99.16} & \underline{97.13} & 99.43 & \underline{99.55} & \textbf{99.18} & \underline{98.77} & \textbf{98.83} & \textbf{98.87} \\
\bottomrule
\end{tabular}
|
83
|
2508.00701
| null |
\caption{
Detection results on 10 VideoPhy datasets.
}
|
\label{tab:videophy}
|
\begin{tabular}{lccccccccccccc}
\toprule
Detection & \multicolumn{10}{c}{Datasets (AP$\uparrow$)} & \multirow{2}{*}{mAP}\\
\cmidrule(r){2-11}
Method & LaVie & OpenSora & CogVideoX-5B & CogVideoX & Dream-Machine & Gen-2 & Pika & SVD & VC2 & ZeroScope \\
\midrule
FID & \underline{96.51} & 87.9 & \underline{91.41} & \underline{93.34} & 97.5 & 98.35 & 99.55 & 95.66 & \underline{96.03} & \underline{90.6} & \underline{94.69} \\
NPR & 63.72 & \underline{88.78} & 81.99 & 81.37 & \textbf{99.86} & \textbf{99.90} & \textbf{99.91} & \textbf{99.54} & 60.21 & 78.23 & 85.35 \\
AIGVDet & 61.06 & 59.07 & 58.95 & 63.15 & 59.27 & 61.55 & 92.96 & 53.73 & 58.22 & 63.11 & 63.11 \\
Demamba & 28.80 & 16.00 & 24.35 & 22.97 & 94.03 & 97.52 & 96.75 & 87.28 & 23.86 & 23.17 & 51.47 \\
Our D3 & \textbf{98.49} & \textbf{98.55} & \textbf{99.03} & \textbf{98.87} & \underline{99.54} & \underline{99.87} & \underline{99.70} & \underline{98.75} & \textbf{99.46} & \textbf{99.38} & \textbf{99.16} \\
\bottomrule
\end{tabular}
|
84
|
2508.00701
| null |
\caption{
Detection results on 6 VidProM datasets.
}
|
\label{tab:vidprom}
|
\begin{tabular}{lcccccccccccccc}
\toprule
Detection & \multicolumn{6}{c}{Datasets (AP$\uparrow$)} & \multirow{2}{*}{mAP}\\
\cmidrule(r){2-7}
Method & MSE & OS & Pika & ST2V & T2VZ & VC2 \\
\midrule
FID & \underline{91.35} & 87.68 & \underline{99.59} & \textbf{97.87} & 68.51 & \underline{85.92} & \textbf{88.49}\\
NPR & 87.04 & \underline{89.85} & \textbf{99.98} & 89.88 & \textbf{88.93} & 70.79 & 87.75\\
AIGVDet & 63.33 & 62.12 & 66.07 & 55.46 & 63.49 & 52.15 & 60.44\\
Demamba & 58.73 & 85.87 & 99.34 & 86.48 & \underline{79.62} & 80.28 & 81.72\\
Our D3 & \textbf{96.85} & \textbf{97.85} & 99.14 & \underline{93.13} & 45.11 & \textbf{98.70} & \underline{88.46}\\
\bottomrule
\end{tabular}
|
85
|
2508.00701
| null |
\caption{Efficiency results on GenVideo with 1000 video samples and batch size of 1. The preprocessing overhead of AIGVDet comes from the optical flow extraction using RAFT.
For image-level methods (FID, NPR), 8 images form a video.}
|
\label{tab:time}
|
\begin{tabular}{l c c c c c}
\toprule
Detection & \multicolumn{3}{c}{Time (s,$\downarrow$)} & mAP$\uparrow$\\
\cmidrule(lr){2-4} \cmidrule(lr){5-5}
Method& Preprocess & Train & Inference & on GenVideo \\
\midrule
FID & Free & 415 & 213 & 88.07\\
NPR & Free & 256 & 188 & 71.26\\
AIGVDet & 500 & 642 & 74 & 70.46\\
Demamba & Free & 196 & 91 & 81.66\\
\midrule
D3 (XCLIP-B/16) & Free & Free & \underline{56} & \textbf{98.46}\\
D3 (MobileNet-v3) & Free & Free & \textbf{40} & \underline{95.47}\\
\bottomrule
\end{tabular}
|
86
|
2508.00701
| null |
\caption{Detection results of D3 against post-processing operations on GenVideo.}
|
\label{tab:robust}
|
\begin{tabular}{ccccccccccccccc}
\toprule
\multirow{2}{*}[-0.5ex]{Visual Encoder} & \multicolumn{5}{c}{Gaussian Blur} & \multicolumn{5}{c}{JEPG Compression} \\
\cmidrule(r){2-6}
\cmidrule(r){7-11}
~ & $\sigma = 0$ & $\sigma = 1$ & $\sigma = 2$ & $\sigma = 3$ & $\sigma = 4$ & $q = 100$ & $q = 90$ & $q = 80$ & $q = 70$ & $q = 60$
\\
\midrule
DINOv2-B & 96.84 & 96.00 & 93.39 & 90.42 & 88.34 & 96.84 & 95.60 & 94.93 & 94.36 & 93.59 \\
DINOv2-L & 96.23 & 95.28 & 92.00 & 88.85 & 87.10 & 96.23 & 94.65 & 93.93 & 93.16 & 92.53 \\
CLIP-B/16 & 97.82 & 97.16 & 83.70 & 83.20 & 84.89 & 97.82 & 84.86 & 83.31 & 81.59 & 78.81 \\
XCLIP-B/16 & \textbf{98.46} & \underline{97.63} & \underline{94.43} & \underline{93.19} & \underline{92.69} & \textbf{98.46} & \underline{97.11} & \underline{96.24} & \underline{95.63} & \underline{94.50} \\
CLIP-B/32 & 97.71 & 97.05 & 84.50 & 85.13 & 86.56 & 97.71 & 91.65 & 89.79 & 88.04 & 86.40 \\
XCLIP-B/32 & \underline{98.15} & \textbf{97.64} & \textbf{95.51} & \textbf{94.22} & \textbf{93.59} & \underline{98.15} & \textbf{97.59} & \textbf{97.37} & \textbf{97.25} & \textbf{96.93} \\
ResNet18 & 97.35 & 96.60 & 92.84 & 90.70 & 90.03 & 97.35 & 96.61 & 95.48 & 94.09 & 92.48 \\
VGG16 & 97.21 & 95.56 & 90.91 & 87.28 & 85.60 & 97.21 & 94.06 & 92.35 & 89.47 & 87.58 \\
EfficientNet-b4 & 96.53 & 95.44 & 90.63 & 88.59 & 88.13 & 96.53 & 94.81 & 93.37 & 91.85 & 90.54 \\
MobileNet-v3 & 96.86 & 95.92 & 87.91 & 84.66 & 84.38 & 96.86 & 94.58 & 91.29 & 88.31 & 85.95 \\
\bottomrule
\end{tabular}
|
96
|
2508.00380
| null |
\caption{Formulations of four numerical functions. Here, $d$ denotes the dimensionality and $\bm{s}$ a random shift in the search space.}
|
\label{tab:numerical}
|
\begin{tabular}{l p{10.5cm}}
\toprule
Function Name & Formulation \\
\midrule
Ackley & $f(\bm{x}) = -a \exp\left(-b \sqrt{\sum_{i=1}^{d} \bm{w}_i^2 / d}\right) - \exp\left(\sum_{i=1}^{d} \cos(2\pi \bm{w}_i) / d\right) + a + \exp(1)$, where $a = 20$, $b = 0.2$, and $\bm{w} = 40\bm{x} - 20 - \bm{s}$ \\
Rosenbrock & $f(\bm{x}) = \sum_{i=1}^{d-1} [100(\bm{w}_{i+1} - \bm{w}_i^2)^2 + (\bm{w}_i - 1)^2]$, where $\bm{w} = 20\bm{x} - 10 - \bm{s}$ \\
Rastrigin & $f(\bm{x}) = 10d + \sum_{i=1}^{d} [\bm{w}_i^2 - 10 \cos(2\pi \bm{w}_i)]$, where $\bm{w} = 64\bm{x} - 32 - \bm{s}$ \\
Levy & $f(\bm{x}) = \sin^2(\pi \bm{w}_1) + \sum_{i=1}^{d-1} [(\bm{w}_i - 1)^2 (1 + 10 \sin^2(\pi \bm{w}_i + 1))] + (\bm{w}_d - 1)^2 (1 + \sin^2(2\pi \bm{w}_d))$, where $\bm{w} = 1 + (20\bm{x} - 10 - \bm{s})/{4}$ \\
\bottomrule
\end{tabular}
|
17,128
|
2501.18324
| null |
\caption{Dialogues created by three separate annotator pairs in VDAct for scenario IDs 00733 and 00912.}
|
\label{tab:example-dials2}
|
\begin{tabular}{c|p{0.28\linewidth}|p{0.28\linewidth}|p{0.28\linewidth}}
\hline \hline
\multicolumn{4}{l}{\textbf{Example}\#3, \textbf{Scenario ID}: 00733, \textbf{Activities}: Turn off light5, Drink wine1, Put groceries in fridge17, Clean stove1, Place wineglass1} \\ \hline
& \textbf{Dialogue ID}: 0073301 & \textbf{Dialogue ID}: 0073302 & \textbf{Dialogue ID}: 0073303 \\ \hline
\multirow{2}{*}{1} & What sorts of things is the man busy doing? & Did the man start in the kitchen? & Where’s the man starting his day? \\ \cline{2-4}
& Drinking, cleaning, putting things away, turning off lights, the usual. & Yes, he did. & He's starting in the kitchen of course. \\ \hline
\multirow{2}{*}{2} & What room does he turn the light off in? & What did he get up to first? & He got any plans in there? \\ \cline{2-4}
& The kitchen. & First, he turned off the light to the kitchen and went into the living room/office. & Actually no as he just turns the lights out and leaves. \\ \hline
\multirow{2}{*}{3} & Had he been busy in the kitchen prior to turning the light off? & What did he do in that part of his house? & Where’s he go to next? \\ \cline{2-4}
& No, he just turned off the light and went to the living room for no apparent reason. & Nothing, he just walked back into the kitchen, where the light was on again, and drank some wine. & He goes to the living room but reverses course immediately and goes back to the kitchen. \\ \hline
\multirow{2}{*}{4} & Is turning off the light the first thing he was seen doing? & How did he drink the wine? & Do you think he forgot something? \\ \cline{2-4}
& Yes. & He drank it mostly from a wine glass but also a little from the bottle too. & Yes he forgot about wine which he decides to drink a ton of. \\ \hline
\multirow{2}{*}{5} & So, after going into the living room he returns to the kitchen? & How long did he spend drinking? & Did he drink it right in the kitchen? \\ \cline{2-4}
& Yes, to have a drink of wine. & About 45 seconds, give or take. & Yes and then he goes to grab a big bag of potato chips. \\ \hline
\multirow{2}{*}{6} & Does he pour himself a glass, or drink it out of the bottle like a savage? & What does he do next? & Does he munch on those as well? \\ \cline{2-4}
& He does both. & Next, he put a bag of potato chips in the fridge. & Again he fakes us all out and just sticks them in the fridge. \\ \hline
\multirow{2}{*}{7} & Are the rest of his activities done in the kitchen? & Where did he get it from originally? & Does he stay in the kitchen after that? \\\cline{2-4}
& Yes, they are. & A high shelf on the wall of the kitchen. & Yea he decides to do some cleaning now. \\ \hline
\multirow{2}{*}{8} & What are they? & What else does he do in the kitchen? & What's he cleaning with? \\ \cline{2-4}
& He puts some chips into the fridge, cleans the stove, and sets out a wine glass, in that order. & He cleans the stove and places a wine glass. & He grabs a sponge and starts working on the stove. \\ \hline
\multirow{4}{*}{9} & Where did the chips come from? & How does he clean the stove? & Do you think he did a good job? \\ \cline{2-4}
& A floating shelf near the range. & Nah but he's already fixated on another wineglass so guess it doesn't matter. & By wiping it with a sponge. \\ \hline
\multirow{2}{*}{10} & What does he clean the stove with? & Where does he place the wineglass? & What's wrong with the wineglass? \\ \cline{2-4}
& A white sponge. & From the TV stand onto the kitchen table. & Nothing I don't think he just decides to move it to the table. \\ \hline \hline
\multicolumn{4}{l}{} \\
\multicolumn{4}{l}{\textbf{Example}\#4, \textbf{Scenario ID}: 00912, \textbf{Activities}: Fall while standing and turning1, Use phone3, Cook fried bread4, Place waterglass1} \\ \hline
& \textbf{Dialogue ID}: 0091201 & \textbf{Dialogue ID}: 0091202 & \textbf{Dialogue ID}: 0091203 \\ \hline
\multirow{2}{*}{1} & What is the man doing when first seen? & What was the man's first activity? & What's the first thing the home skillet does this fine day? \\ \cline{2-4}
& He is seen going into the living room, picking up the TV remote, and as he is turning around tripping or slipping and falling down. & He had a fall in the living room. & He begins in the kitchen. \\ \hline
\multirow{2}{*}{2} & Okay, does he resume doing something in the living room after the fall? & How did that happen? & What's he got going on there? \\ \cline{2-4}
& Yes, he picks up his cell phone. & Not entirely sure, he was just walking when he fell forward. & He isn't there for long as he starts wandering around the house. \\ \hline
\multirow{2}{*}{3} & Does he sit down on the sofa to use it? & Was he able to move on to his next activity? & How many rooms does he wonder to? \\ \cline{2-4}
& Yes, but only after returning from using it in the bedroom. & Yes, he was. & About two. He finally arrives in the living room to which he goes over and turns on nightmare bunny. \\ \hline
\multirow{2}{*}{4} & Where did the man get the phone? & What was his next activity? & Why would he do that again? \\ \cline{2-4}
& From the coffee table in the living room. & He used his phone a couple of times. & Bored. After that he grabs the remote and attempts to sit down with it. \\ \hline
\multirow{2}{*}{5} & Where does he go after using the phone on the sofa? & Where did he pick up his phone from? & Does he not succeed? \\ \cline{2-4}
& To the kitchen. & It was next to the TV in the living room. & Nope, he falls down face first. \\ \hline
\multirow{2}{*}{6} & What does his visit to the kitchen consist of? & Where did he use it? & Did he hit his head on anything? \\ \cline{2-4}
& He does some cooking and some relocation of two water glasses. & In the bedroom and in the living room. & He seems fine, now he grabs his cellphone and takes it to his bedroom where he plays with it there on the bed. \\ \hline
\multirow{2}{*}{7} & Does he cook or handle the glasses first? & What did he do after using his phone? & What do you think he is playing? \\ \cline{2-4}
& He does the cooking first. & He went to the kitchen to do some cooking. & Tetris as per usual. He only spends a couple of seconds with it on the bed and after that he takes it back to the couch to play with it there too. \\ \hline
\multirow{2}{*}{8} & What is the man cooking? & What did he cook? & Does he continue playing Tetris? \\ \cline{2-4}
& Some French toast. & Some fried bread with milk. & No. After that he gets real hungry so he goes to the kitchen to make his famous burnt toast and milk on the stove. \\ \hline
\multirow{2}{*}{9} & How many ingredients are used for the dish? & Did he cook it on the stove? & Does he finally enjoy it? \\ \cline{2-4}
& Two, toast and milk. & Yes, he did. & Maybe he did but the next thing I know is he's getting two water glasses. \\ \hline
\multirow{2}{*}{10} & From where does the man relocate the water to and from? & What did he do last? & Does he double fist drinking water? \\ \cline{2-4}
& From two different spots on the kitchen table to another place on the same table. & He placed some water glasses on the kitchen table. & All he does is pick them up and then set them on the table. \\ \hline
\end{tabular}
|
87
|
2508.00701
| null |
\caption{
Ablation studies of visual encoder backbones and the type of first-order features (L2 Distance or Cosine Similarity).
}
|
\label{tab:ablation}
|
\begin{tabular}{lcccccccc}
\toprule
\multirow{2}{*}[-0.5ex]{Visual Encoder} & \multicolumn{2}{c}{GenVideo} & \multicolumn{2}{c}{EvalCrafter} & \multicolumn{2}{c}{VideoPhy} & \multicolumn{2}{c}{VidProM} \\
\cmidrule(r){2-3} \cmidrule(r){4-5} \cmidrule(r){6-7} \cmidrule(r){8-9}
~ & L2 & Cos & L2 & Cos & L2 & Cos & L2 & Cos \\
\midrule
DINOv2-B & 95.84 & 87.17 & 96.76 & 89.31 & 93.98 & 82.14 & 82.17 & 73.23 \\
DINOv2-L & 94.92 & 85.33 & 95.84 & 87.31 & 92.49 & 79.12 & 80.90 & 70.83 \\
CLIP-B/16 & \underline{97.00} & 87.82 & 97.63 & 89.82 & 97.01 & 86.24 & 84.79 & 75.77 \\
XCLIP-B/16 & \textbf{97.72} & \underline{91.30} & \textbf{98.24} & \underline{92.81} & \underline{97.14} & \underline{89.10} & \textbf{87.08} & \textbf{79.87} \\
CLIP-B/32 & 96.73 & 87.87 & 97.26 & 89.53 & 96.61 & 87.04 & 83.97 & 75.52 \\
XCLIP-B/32 & 96.99 & 90.43 & \underline{97.72} & 92.31 & 96.35 & 88.74 & \underline{85.57} & \underline{79.62} \\
ResNet-18 & 96.39 & 89.73 & 97.26 & 91.64 & 95.67 & 86.83 & 81.59 & 75.68 \\
VGG-16 & 96.97 & \textbf{92.63} & \underline{97.84} & \textbf{94.16} & \textbf{97.50} & \textbf{91.21} & 81.54 & 77.02 \\
EfficientNet-B4 & 94.28 & 85.51 & 95.49 & 88.08 & 92.46 & 82.40 & 80.73 & 73.00 \\
MobileNet-V3 & 95.47 & 87.14 & 96.48 & 89.50 & 94.70 & 84.71 & 80.76 & 73.74 \\
\bottomrule
\end{tabular}
|
88
|
2508.00701
| null |
\caption{Ablation studies of feature order on 4 datasets.}
|
\label{tab:order}
|
\begin{tabular}{l c c c c c}
\toprule
Detection & \multicolumn{2}{c}{GenVideo} & \multicolumn{2}{c}{EvalCrafter}\\
\cmidrule(r){2-3} \cmidrule(r){4-5}
Method & mAP$\uparrow$ & Avg. AUC$\uparrow$ & mAP$\uparrow$ & Avg. AUC$\uparrow$ \\
\midrule
D3 (1st-Order) & 95.69 & 93.45 & 86.40 & 85.17 \\
D3 (2nd-Order) & \textbf{98.46} & \textbf{97.72} & \textbf{98.87} & \textbf{98.24} \\
\midrule
~ & \multicolumn{2}{c}{VideoPhy} & \multicolumn{2}{c}{VidProM}\\
\cmidrule(r){2-3} \cmidrule(r){4-5}
~ & mAP$\uparrow$ & Avg. AUC$\uparrow$ & mAP$\uparrow$ & Avg. AUC$\uparrow$ \\
\midrule
D3 (1st-Order) & 86.06 & 84.22 & 80.61 & 77.31 \\
D3 (2nd-Order) & \textbf{99.16} & \textbf{97.14} & \textbf{88.46} & \textbf{87.08} \\
\bottomrule
\end{tabular}
|
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 17