multiple \includegraphics in same line, only first is recognized, plots are missed

[tsgit]

http://arxiv.org/e-print/arXiv:1801.10178
has a TeX file with

\begin{figure*}[t]                                                                                                                                                  
\centering                                                                                                                                                          
\includegraphics[width=0.5\textwidth]{ps_CO_z6_fnl6p5.pdf}\includegraphics[width=0.5\textwidth]{ps_CII_z2p8_fnl6p5.pdf}                                             
\caption{The spherically averaged clustering component of the power spectrum (including AP effect) for $f_{\rm NL}^{\rm loc} = 6.5$ (solid blue) and $f_{\rm NL}^{\\
rm loc} = 1$ (dashed magenta) for CO(1-0) at $z\simeq 6$ (left) and [CII] (right) at $z\simeq 2$ are shown. These redshifts are chosen since for the experiments co\
nsidered here, they provide the tightest constraints on $f_{\rm NL}^{\rm loc}$. The shaded blue region is the expected spherically averaged variance for the power \
spectrum with $f_{\rm NL}^{\rm loc} = 6.5$, defined in Eq. (\ref{eq:error}). For illustration we also show the spherically averaged Gaussian contribution (dashed-d\
otted purple), as well as the shot contribution (dotted red).  The vertical lines correspond to the smallest scale $k_{\rm max}$ considered in our forecast, the ch\
oice of which is described in the text.                                                                                                                             
\label{Fig:PS}}                                                                                                                                                     
\end{figure*}                                                                                                                                                       

so there are 2 different includegraphics commands on the same line. The code misses the second one.
Specifically

In [3]: extract_captions("./paper_v2.tex", ".", ["./ps_CII_z2p8_fnl6p5.pdf", "./ps_CO_z6_fnl6p5.pdf"])
Out[3]: 
[('ps_CO_z6_fnl6p5.pdf',
  'The spherically averaged clustering component of the power spectrum (including AP effect) for $f_{\\rm NL}^{\\rm loc} = 6.5$ (solid blue) and $f_{\\rm NL}^{\\rm loc} = 1$ (dashed magenta) for CO(1-0) at $z\\simeq 6$ (left) and [CII] (right) at $z\\simeq 2$ are shown. These redshifts are chosen since for the experiments considered here, they provide the tightest constraints on $f_{\\rm NL}^{\\rm loc}$. The shaded blue region is the expected spherically averaged variance for the power spectrum with $f_{\\rm NL}^{\\rm loc} = 6.5$, defined in Eq. (\\ref{eq:error}). For illustration we also show the spherically averaged Gaussian contribution (dashed-dotted purple), as well as the shot contribution (dotted red). The vertical lines correspond to the smallest scale $k_{\\rm max}$ considered in our forecast, the choice of which is described in the text.',
  'Fig:PS')]

when a newline is inserted, the result is instead:

In [4]: extract_captions("./paper_v2.tex", ".", ["./ps_CII_z2p8_fnl6p5.pdf", "./ps_CO_z6_fnl6p5.pdf"])
Out[4]: 
[('ps_CO_z6_fnl6p5.pdf',
  'The spherically averaged clustering component of the power spectrum (including AP effect) for $f_{\\rm NL}^{\\rm loc} = 6.5$ (solid blue) and $f_{\\rm NL}^{\\rm loc} = 1$ (dashed magenta) for CO(1-0) at $z\\simeq 6$ (left) and [CII] (right) at $z\\simeq 2$ are shown. These redshifts are chosen since for the experiments considered here, they provide the tightest constraints on $f_{\\rm NL}^{\\rm loc}$. The shaded blue region is the expected spherically averaged variance for the power spectrum with $f_{\\rm NL}^{\\rm loc} = 6.5$, defined in Eq. (\\ref{eq:error}). For illustration we also show the spherically averaged Gaussian contribution (dashed-dotted purple), as well as the shot contribution (dotted red). The vertical lines correspond to the smallest scale $k_{\\rm max}$ considered in our forecast, the choice of which is described in the text.',
  'Fig:PS'),
 ('ps_CII_z2p8_fnl6p5.pdf',
  'The spherically averaged clustering component of the power spectrum (including AP effect) for $f_{\\rm NL}^{\\rm loc} = 6.5$ (solid blue) and $f_{\\rm NL}^{\\rm loc} = 1$ (dashed magenta) for CO(1-0) at $z\\simeq 6$ (left) and [CII] (right) at $z\\simeq 2$ are shown. These redshifts are chosen since for the experiments considered here, they provide the tightest constraints on $f_{\\rm NL}^{\\rm loc}$. The shaded blue region is the expected spherically averaged variance for the power spectrum with $f_{\\rm NL}^{\\rm loc} = 6.5$, defined in Eq. (\\ref{eq:error}). For illustration we also show the spherically averaged Gaussian contribution (dashed-dotted purple), as well as the shot contribution (dotted red). The vertical lines correspond to the smallest scale $k_{\\rm max}$ considered in our forecast, the choice of which is described in the text.',
  'Fig:PS')]