image_colour.tex

\documentclass[12pt]{beamer}

\title[Image-Colourization]{Image Colourisation Project}
\subtitle{Jun-Aug 2020 {\textbar} WTEF Project {\textbar} Deep Learning}
\author[Team-9]{Priyansi {\textbar} Sejal Gupta {\textbar} Twisha Bansal}
\date{31 August 2020}

\usetheme{metropolis}
\usepackage{xcolor}
\usepackage{graphicx}
\usepackage{hyperref}

\graphicspath{ {/mnt/c/Users/Sejal/Desktop/project/} }

\begin{document}

\begin{frame}
        \titlepage
\end{frame}

\begin{frame}{Objective}
	\begin{center}
		\includegraphics[width=11cm]{Objective}
	\end{center}
\end{frame}

\begin{frame}{Motivation}
	\begin{center}
		\includegraphics[width=10cm]{slide_fascination}
	\end{center}
\end{frame}

\begin{frame}[standout]
	\center{Our journey from being entirely clueless to completing an\\ Image Colourisation project\\ in Deep Learning}
\end{frame}

\begin{frame}{Colorful Image Colorization paper by Richard Zhang, Phillip Isola, Alexei A. Efros}
	\textbf{To hallucinate the most plausible colour version rather than the ground truth}\\
	\vspace{10 mm}
	\pause
	\textbf{Train a CNN to map from a grayscale input to a distribution over quantized
	colour value outputs}
\end{frame}

\begin{frame}{Colorful Image Colorization paper by Richard Zhang, Phillip Isola, Alexei A. Efros}
	\textbf{Challenges}\\
	\begin{itemize}
		\item Advanced Mathematics
		\item Obtaining the ImageNet Dataset
		\item Availability of a GPU
		\item Uploading a 150 GB dataset online
	\end{itemize}
\end{frame}

\begin{frame}[standout]
	\center{\alert{AutoEncoders}}\\
	A type of Neural Network used to learn representation for a set of data in an unsupervised manner
\end{frame}

\begin{frame}{CIELAB Colour Space}
	\begin{itemize}
		\item Why RGB will not work\\~
	\end{itemize}
\end{frame}

\begin{frame}[standout]
	\begin{columns}
		\begin{column}{0.5\textwidth}
			\center{\alert{\textbf{Grayscale}}}\\
			\center{Only 1 channel}
		\end{column}
		\begin{column}{0.5\textwidth}
			\center{\alert{\textbf{RGB}}}\\
			\center{3 channels}
		\end{column}
	\end{columns}
	\pause
	\vspace{5 mm}
	\center{\alert{\textbf{Grayscale $\Rightarrow$ RGB}}}
	\center{3 channels, but R==G==B}
\end{frame}

{\setbeamercolor{background canvas}{bg=gray}
	\begin{frame}
	\end{frame}
}

\begin{frame}{CIELAB Colour Space}
	\begin{itemize}
		\item L channel: Lightness
		\item A channel: green to red
		\item B channel: blue to yellow
	\end{itemize}
\end{frame}

\begin{frame}{Technology Stack}
	\begin{columns}
		\begin{column}{0.5\textwidth}
			\begin{itemize}
				\item \textbf{Building the Model}\\
					- PyTorch\\
				\item \textbf{Datasets and Version Control}\\
					- Kaggle\\
			\end{itemize}
		\end{column}
		\begin{column}{0.5\textwidth}
			\begin{itemize}
				\item \textbf{Web App}\\
					- Streamlit\\
				\item \textbf{Cloud Platform}\\
					- Heroku
			\end{itemize}
		\end{column}
	\end{columns}
\end{frame}

\begin{frame}{Model}
	\begin{center}
		\includegraphics[width=11cm]{Model}
	\end{center}
\end{frame}

\begin{frame}{Model}
	\begin{itemize}
		\item Loss Function: \textbf{MSE Loss}
		\item Optimiser: \textbf{Adam}
		\item Range of Learning Rates used for training: \textbf{1e-3 - 1e-6}
	\end{itemize}
\end{frame}

\begin{frame}{Challenges}
	\begin{itemize}
		\item Tensor and Numpy Array interconvertability
		\item Interoperatability between CPU and CUDA
		\item Runtime disconnects
		\item Signal being killed due to memory usage
			\pause
		\item No Mathematical Parameters like Accuracy to check results
	\end{itemize}
\end{frame}

\begin{frame}{Datasets}
	Trained on approximately 313K images
	\begin{itemize}
		\item \href{https://www.kaggle.com/lijiyu/imagenet}{ImageNet(50K Images)}\\
		\item \href{https://www.kaggle.com/hsankesara/flickr-image-dataset}{Flickr}\\
		\item \href{https://www.kaggle.com/huseynguliyev/landscape-classification}{Landscape Classification}\\
		\item \href{https://www.kaggle.com/nitishabharathi/scene-classification}{Scene Classification}\\
		\item \href{https://www.kaggle.com/moltean/fruits}{Fruits 360}\\
		\item \href{https://www.kaggle.com/chrisfilo/fruit-recognition}{Fruits Recognition}\\
		\item \href{https://www.kaggle.com/salil007/caavo}{Clothes Classification}\\
	\end{itemize}
\end{frame}

\begin{frame}{Results: Pass}
	\includegraphics[width=11cm]{pass}
\end{frame}

\begin{frame}{Results: Fail}
	\includegraphics[width=11cm]{fail}
\end{frame}

\begin{frame}{Web Application and Deployment}
	\textbf{Web App:}\\
	\begin{itemize}
		\item Made with Streamlit\\
		\item Why Streamlit:\\
			\begin{enumerate}
				\item Reducing app code to Python scripts
				\item Treating widgets like variables
				\item Reusing data with memoization
			\end{enumerate}
	\end{itemize}
	\textbf{Deployment:}
	\begin{itemize}
		\item Heroku as the cloud platform
	\end{itemize}
\end{frame}

\begin{frame}{Further Improvements}
	\begin{itemize}
		\item \textbf{Model}
			\begin{enumerate}
				\item Generalising on more themes
				\item Automating the classification of themes
				\item Incorporating Data augmentation
				\item Compatibility with all image sizes
			\end{enumerate}
		\item \textbf{Web App}
			\begin{enumerate}
				\item Using Flask along with HTML/CSS/JS to customize better
			\end{enumerate}
	\end{itemize}
\end{frame}

\begin{frame}{References}
	\begin{enumerate}
		\item Colorful Image Colorization paper by Richard Zhang, Phillip Isola, Alexei A. Efros: \textcolor{cyan}{\href{https://arxiv.org/pdf/1603.08511.pdf}{https://arxiv.org/pdf/1603.08511.pdf}}
		\item Applications of AutoEncoders - Image Colourisation: \textcolor{cyan}{\href{https://github.com/bnsreenu/python\_for\_microscopists}{https://github.com/bnsreenu/python\_for\_microscopists}}
	\end{enumerate}
\end{frame}

\begin{frame}{Our Project}
	\begin{itemize}
		\item Web Application: \textcolor{cyan}{\href{https://image-colouriser-streamlit.herokuapp.com/}{https://image-colouriser-streamlit.herokuapp.com/}}
		\item Gitlab: \textcolor{cyan}{\href{https://gitlab.com/twishabansal/image-colourisation}{https://gitlab.com/twishabansal/image-colourisation}}
		\item Kaggle Notebook: \textcolor{cyan}{\href{https://www.kaggle.com/sejalgupta01/image-colorization-starter}{https://www.kaggle.com/sejalgupta01/image-colorization-starter}}
	\end{itemize}
\end{frame}

\begin{frame}[standout]
	\center{Questions and Suggestions?}
\end{frame}

\end{document}