Table/Figure title changes.

chapters/01_introduction.tex

@@ -23,7 +23,7 @@ \section{Problem Statement}
 \begin{figure}[htpb]
 	\centering
 	\includegraphics[width=1.0\linewidth]{figures/ucf-intro/serie1.png} 
-	\caption[Example Image Sequence]{Example of an image sequence with an unknown future frame. The sequence is starting from the left and is taken from UCF-101.} \label{fig:intro-seq}
+	\caption[Image Sequence Example]{Example of an image sequence with an unknown future frame. The sequence is starting from the left and is taken from UCF-101.} \label{fig:intro-seq}
 \end{figure}
 
 Like in the first example of object recognition on static pictures, this task might sound trivial for humans once more, since we already have built an intuition regarding motion and our environment. When we have a look at the image sequence in Figure \ref{fig:intro-seq}, we have a strong idea about how this sequence might continue. At least for a couple of time steps. The boy in the foreground probably will lift his left foot towards the ball, while the ball continues to fall down due to gravitation. In contrast, the background will stay almost unchanged.

chapters/02_fundamentals.tex

@@ -409,7 +409,7 @@ \subsubsection{Structure}
   \caption{}
   \label{fig:rnn-many2many}
 \end{subfigure}
-\caption[Examples of RNN Input-Output Modes]{Visualization of different recurrent network input-output modes: (a) one-to-many, (b) many-to-one, (c) many-to-many. Red squares denote the inputs, gray squares the recurrent cells and all outputs are colored in blue. Input and output squares can be understood as either further neural networks or the direct input and output. (Based on \parencite{rnn-effectiveness})}
+\caption[RNN Input-Output Modes]{Visualization of different recurrent network input-output modes: (a) one-to-many, (b) many-to-one, (c) many-to-many. Red squares denote the inputs, gray squares the recurrent cells and all outputs are colored in blue. Input and output squares can be understood as either further neural networks or the direct input and output. (Based on \parencite{rnn-effectiveness})}
 \label{fig:rnn-modes}
 \end{figure}
 

chapters/06_evaluation.tex

@@ -559,7 +559,7 @@ \subsubsection{Quantitative Results}
       3enc-ConvLSTM-SS-3dec(5/64-64-64) & \num{5299841} & \textbf{0.0407} \\
     \bottomrule
   \end{tabular}
-  \caption[Results on Moving MNIST]{Comparison with other networks on Moving MNIST. The numbers in brackets identify the \textit{amount of hidden units} per layer in case of FC-LSTMs, and for ConvLSTMs the \textit{hidden-to-hidden kernel size} followed by the \textit{feature maps per layer} are listed. Further, \textit{3enc} denotes tree convolutional layers in the spatial encoder. We call our model 3enc-ConvLSTM-SS-3dec.}\label{tab:mm-comparison}
+  \caption[Test Results on Moving MNIST]{Comparison with other networks on Moving MNIST. The numbers in brackets identify the \textit{amount of hidden units} per layer in case of FC-LSTMs, and for ConvLSTMs the \textit{hidden-to-hidden kernel size} followed by the \textit{feature maps per layer} are listed. Further, \textit{3enc} denotes tree convolutional layers in the spatial encoder. We call our model 3enc-ConvLSTM-SS-3dec.}\label{tab:mm-comparison}
 \end{table}
 
 \subsubsection{Qualitative Results}
@@ -903,7 +903,7 @@ \subsubsection{Quantitative Results}
       MSE+GDL+SSIM & \textbf{0.0023} & \textbf{0.0005} & \textbf{0.0037} & \textbf{0.0014} \\
     \bottomrule
   \end{tabular}
-  \caption[Test Errors on MsPacman]{Absolute and squared error test results on MsPacman dataset using varying loss layers in our 2-layer network.}\label{tab:pac-comparison2}
+  \caption[Test Results on MsPacman]{Absolute and squared error test results on MsPacman dataset using varying loss layers in our 2-layer network.}\label{tab:pac-comparison2}
 \end{table}