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
2024-06-20 13:08:28 +02:00
parent 0969f767e9
commit 0f10df50fb
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@ -164,7 +164,7 @@ AlexNet is composed of:
\subsection{Training} \subsection{Training}
The network is trained on ImageNet 1k (for the ILSVRC 2012) with a batch size of 128. The network was trained on ImageNet 1k (for the ILSVRC 2012) with a batch size of 128.
Due to GPU memory limitations, training was split into two parallel lines on two GPUs (model parallelism). Due to GPU memory limitations, training was split into two parallel lines on two GPUs (model parallelism).
\begin{description} \begin{description}
@ -675,7 +675,7 @@ It has the following properties:
Due to the shape mismatch, the output of a stage cannot be directly used as the skip connection of the next stage. Due to the shape mismatch, the output of a stage cannot be directly used as the skip connection of the next stage.
Possible solutions are: Possible solutions are:
\begin{itemize} \begin{itemize}
\item Apply stride $2$ and zero-padding in the first layer of the next stage (this does not add new parameters). \item Apply stride $2$ and zero-pad the missing channels of the skip connection (this does not add new parameters).
\item The output of the previous stage is passed through a $1 \times 1$ convolution with stride $2$ and $2C$ output channels (shown to work slightly better). \item The output of the previous stage is passed through a $1 \times 1$ convolution with stride $2$ and $2C$ output channels (shown to work slightly better).
\end{itemize} \end{itemize}