notxia/unibo-ai-notes
1
0
Fork 0
mirror of https://github.com/NotXia/unibo-ai-notes.git synced 2025-12-15 19:12:22 +01:00
Code Issues Packages Projects Releases Wiki Activity

Fix typos <noupdate>

Browse Source
This commit is contained in:
Tian Cheng (Riccardo) Xia
2024-06-24 20:26:41 +02:00
parent 0dbb1068ad
commit ed95636174
2 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/year1/image-processing-and-computer-vision/module2/sections/_architectures.tex
View File

@ -599,7 +599,7 @@ Different modules are used depending on the activation shape.
\subsection{Inception-v4} \subsection{Inception-v4}
\marginnote{Inception-v4} \marginnote{Inception-v4}
A larger version of Inception v3 with more complicated stem layers. A larger version of Inception-v3 with more complicated stem layers.
@ -660,7 +660,7 @@ It has the following properties:
\item A stage is composed of residual blocks. \item A stage is composed of residual blocks.
\item A residual block is composed of two $3 \times 3$ convolutions followed by batch normalization. \item A residual block is composed of two $3 \times 3$ convolutions followed by batch normalization.
\item The first residual block of each stage halves the spatial dimension and doubles the number of channels (there is no pooling). \item The first residual block of each stage halves the spatial dimension and doubles the number of channels (there is no pooling).
\item Stem layers are less aggressive than GoogLeNet (\texttt{conv + pool}. Input reduced to $56 \times 56$). \item Stem layers are less aggressive than GoogLeNet (\texttt{conv + pool}. Input reduced to a shape of $56 \times 56$).
\item Global average pooling is used instead of flattening. \item Global average pooling is used instead of flattening.
\end{itemize} \end{itemize}
@ -696,7 +696,7 @@ It has the following properties:
\end{description} \end{description}
\begin{remark} \begin{remark}
ResNet improves the results of a deeper layer but beyond a certain depth, the gain is negligible. ResNet improves the results of a deeper network but, beyond a certain depth, the gain is negligible.
\begin{figure}[H] \begin{figure}[H]
\centering \centering
\includegraphics[width=0.65\linewidth]{./img/resnet_results.png} \includegraphics[width=0.65\linewidth]{./img/resnet_results.png}

4
src/year1/image-processing-and-computer-vision/module2/sections/_image_formation.tex
View File

@ -621,7 +621,7 @@ Therefore, the complete workflow for image formation becomes the following:
\item[Homographies non-linear refinement] \item[Homographies non-linear refinement]
The homographies $\matr{H}_i$ estimated at the previous step are obtained using a linear method and need to be refined as, for each image $i$, The homographies $\matr{H}_i$ estimated at the previous step are obtained using a linear method and need to be refined as, for each image $i$,
the IRF coordinates $\matr{H}_i\vec{w}_j = \left( \frac{h_{i, 1}^T \tilde{\vec{w}}_j}{h_{i, 3}^T \tilde{\vec{w}}_j}, \frac{h_{i, 2}^T \tilde{\vec{w}}_j}{h_{i, 3}^T \tilde{\vec{w}}_j} \right)$ the IRF coordinates $\matr{H}_i\vec{w}_j = \left( \frac{h_{i, 1}^T \tilde{\vec{w}}_j}{h_{i, 3}^T \tilde{\vec{w}}_j}, \frac{h_{i, 2}^T \tilde{\vec{w}}_j}{h_{i, 3}^T \tilde{\vec{w}}_j} \right)$
of the world point $\vec{w}_j$ are still not matching the known IRF coordinates $\vec{m}_{i,j}$ of the $j$-corner in the $i$-image. of the world point $\vec{w}_j$ are still not matching the known IRF coordinates $\vec{m}_{i,j}$ of the $j$-th corner in the $i$-th image.
\begin{figure}[H] \begin{figure}[H]
\centering \centering
\includegraphics[width=0.7\linewidth]{./img/_homography_refinement.pdf} \includegraphics[width=0.7\linewidth]{./img/_homography_refinement.pdf}
@ -989,7 +989,7 @@ Undistorted images enjoy some properties:
\begin{example}[Compensate pitch or yaw] \begin{example}[Compensate pitch or yaw]
In autonomous driving, cameras should be ideally mounted with the optical axis parallel to the road plane and aligned with the direction of motion. In autonomous driving, cameras should be ideally mounted with the optical axis parallel to the road plane and aligned with the direction of motion.
It is usually very difficult to obtain perfect alignment physically It is usually very difficult to physically obtain perfect alignment
but a calibrated camera can help to compensate pitch (i.e. rotation around the $x$-axis) but a calibrated camera can help to compensate pitch (i.e. rotation around the $x$-axis)
and yaw (i.e. rotation around the $y$-axis) by estimating the vanishing point of the lane lines. and yaw (i.e. rotation around the $y$-axis) by estimating the vanishing point of the lane lines.