Untitled


<html>

<head>

<title>Computer Vision Mid-Term Exam 01</title>


<link href='http://fonts.googleapis.com/css?family=Nunito:300|Crimson+Text|Droid+Sans+Mono|Dancing+Script' rel='stylesheet' type='text/css'>


<script src="http://ajax.googleapis.com/ajax/libs/jquery/1.3.2/jquery.min.js"></script>


<link rel="stylesheet" href="highlight/styles/default.css">

<script src="highlight/highlight.pack.js"></script>

<script>hljs.initHighlightingOnLoad();</script>


<script type="text/x-mathjax-config">

MathJax.Hub.Config({

  tex2jax: {inlineMath: [['$','$'], ['\\(','\\)']]}

});

</script>

<script src='https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML'></script>


<link rel="stylesheet" href="styles.css">


</head>

<body>


<div id="header"><div id="headersub">

<h1>Elise White</h1>

</div></div>


<div class="container">


<h2>COS 351 Mid-Term Exam 01</h2>


<p>

Out: Saturday, 2016 April 30, 9:00pm<br>

Due: Monday, 2016 May 9, 11:55pm

</p>


<h3 id="instructions">Instructions</h3>


<div class="instructions">


<p>This mid-term exam will be done outside the classroom.  This exam will ask questions that are generally at a higher level and more synthesizing than the problems of the regular assignments.  You will be asked to write some <em>working</em> MATLAB/Octave code, write short paragraph answers, generate images, etc.</p>


<p>I strongly encourage you to show all of your work.  If you used MATLAB/Octave to perform computations, copy-paste your log in <code>&lt;pre>&lt;code></code> tags.  I also recommend inserting your images of your work captured by screenshot, camera, or scanner. If you want to include math, you are encouraged to use the included <a href="https://www.mathjax.org/" target="_blank">MathJax</a> by wrapping the math in single dollar signs for inline math (<code>$f(x) = 4x^2 + x - 2$</code> produces $f(x) = 4x^2 + x - 2$) or double dollar sign for display math (<code>$$f(x) = 4x^2 + x - 2$$</code> produces below)</p>


$$f(x) = 4x^2 + x - 2$$


<p>Be thorough in your explanation.  This work is relatively easy to understand conceptually but much more challenging to implement and explain.  Your job is to convince me that you understand the problem and your solution.  A convincing argument may require a detailed explanation.  However, be careful not to ramble either.  Be clear and concise.  It is perfectly fine to admit that you are not sure of your answer.  It is not fine to convince yourself that you understand the problem, when it is clear by your solution that you do not.</p>


<p>Each question may have several parts.  Be sure to answer each part, and it would help to indicate each part's solution with some mark, such as a (a), or wrap your answers with <code>&lt;ol class="answers"></code>.</p>


<p>Be sure to zip and upload <em>all</em> of the files necessary to view your submission correctly!  For example, if you add an image to this document, be sure that the image file is located in the same folder, and that the image is included in the final zip.</p>


<p>You are allowed to use the following:</p>


<ul>

<li>your own notes, assignment solutions, assignment write ups</li>

<li>slide sets, course book</li>

<li>web to search for MATLAB/Octave commands</li>

</ul>


<p>You are <em>not allowed</em> to do any of the following:</p>


<ul>

<li>use other student's notes or assignment solutions</li>

<li>talk with other students about any part of the problems (including clarifying questions!)</li>

<li>use the Internet to look up terms or concepts</li>

</ul>


<p>Hint: Run the code in MATLAB/Octave.  You may use the provided images, or you may use your own.</p>


<p>If you have any questions, you may email me or post <em>privately</em> to Piazza.</p>


</div>


<!--  ========================================================  -->

<!--  ========================================================  -->


<h3 id="h1">q1. Computer Vision (9pts)</h3>


<div class="questions" id="q1">

<p>The following is a comic by Randall Munroe of XKCD (<a href="http://xkcd.com/1425/" target="_blank">source</a>).</p>


<figure><img src="xkcd-1425-tasks.png" title="In the 60s, Marvin Minsky assigned a couple of undergrads to spend the summer programming a computer to use a camera to identify objects in a scene. He figured they'd have the problem solved by the end of the summer. Half a century later, we're still working on it." alt="Tasks"></figure>


<p>The <code>title</code> attribute of the comic contains the following text.</p>


<blockquote>In the 60s, Marvin Minsky assigned a couple of undergrads to spend the summer programming a computer to use a camera to identify objects in a scene. He figured they'd have the problem solved by the end of the summer. Half a century later, we're still working on it.</blockquote>


<ol class="questions">

<li>Explain the joke in the character's conversation.</li>

<li>Explain the joke found in the caption ("In CS, it can be hard...").</li>

<li>Marvin Minsky recently passed away.  He won many awards during his career.  Report something interesting about his career, his personal life, or his views of machines and humans.</li>

</ol>


</div>


<div class="answers" id="a1">

<p>Write your answers below.</p>

<ol class="answers">

<li>The joke is that it seems like it would be easy to recognize a bird in an image, as it's something that humans can do instantaneously. However, it takes hours and hours of development for a computer to do the same thing.</li>

<li>As humans, it's easy for us to recognize birds. But if we were placed in a random location in the US, depending on the situation it could be nearly impossible for us to tell whether we were in a national park or not. In CS, things that seem like they would be easy are very difficult, and things that seem difficult are very easy.
</li>

<li>In 1951, Minsky built the first neural network simulator (SNARC). I learned that an artificial neural network is a system that simulates the way that neurons connect in brains.</li>

</ol>

</div>


<!--  ========================================================  -->

<!--  ========================================================  -->


<h3 id="h2">q2. Machine Learning (9pts)</h3>


<div class="questions" id="q2">

<p>In class, I briefly how generalization is important to classification and categorization.</p>


<ol class="questions">

<li>Explain what "generalization" means with respect to the model learned.</li>

<li>What is the cause of bias error?  How can you eliminate it?  Is that something you want to do?  Why or why not?</li>

<li>What is variance?  Can you eliminate it?  If so, how?  If not, why not, and how far can you reduce it?</li>

</ol>

</div>


<div class="answers" id="a2">

<ol class="answers">

<li>(write your answer here)</li>

<li>(write your answer here)</li>

<li>(write your answer here)</li>

</ol>

</div>


<!--  ========================================================  -->

<!--  ========================================================  -->


<h3 id="h3">q3. 3D Reconstruction / Structure from Motion (9pts)</h3>


<div class="questions" id="q3">


<p>We discussed in class all of the parts necessary to reconstruct a 3D scene from a set of images.</p>


<ol class="questions">

<li>Describe in summary what the entire system would look like.  For example, what is the input to the system, the output of the system, what subsystems are within, the inputs and outputs of each subsystem, etc.  Your system should be complete (not missing a subsystem).</li>

<li>Explain why each subsystem is important to the final system.</li>

<li>Detail the limitations of each subsystem, and how you would mitigate or work around these limitations if you were to implement it.  For example, explain what assumptions each subsystem makes about its input.</li>

</ol>

</div>


<div class="answers" id="a3">

<ol class="answers">

<li>(write your answer here)</li>

<li>(write your answer here)</li>

<li>(write your answer here)</li>

</ol>

</div>


<!--  ========================================================  -->

<!--  ========================================================  -->


<h3 id="h4">q4. 3D Reconstruction Limitations (9pts)</h3>


<div class="questions" id="q4">

<p>For the following configurations, explain why or why not the 3D scene can be fully reconstructed.  If the scene isn't reconstructable, explain what can be done with the images</p>


<figure>

    <img src="camera-center.png" alt="Cameras with same optical center" height="150">

    <img src="satellite-aerial-photos-of-earth-1.jpg" alt="Bourtange, Vlagtwedde, Netherlands" height="150">

    <img src="camera-right.png" alt="Cameras with optical lines at right angles." height="150">

    <figcaption>Satellite image <a href="http://www.boredpanda.com/daily-overview-satellite-aerial-photography-earth/" target="_blank">source</a>.</figcaption>

</figure>


<ol class="questions">

<li>Two images taken by a single camera that has been rotated about its optical center, where fields of view overlap. (left subfigure)</li>

<li>Two images taken by cameras with telephoto lenses.  For example, satellite images, such as the figure at right. (center subfigure)</li>

<li>Two images taken at right angles to each other. (right subfigure)</li>

</ol>

</div>


<div class="answers" id="a4">

<ol class="answers">

<li>(write your answer here)</li>

<li>(write your answer here)</li>

<li>(write your answer here)</li>

</ol>

</div>


<!--  ========================================================  -->

<!--  ========================================================  -->


<h3 id="h5">q5. Model Fitting (9pts)</h3>


<div class="questions" id="q5">

<p>Use the following code to generate and visualize 100 data points.</p>


<pre><code class="matlab">x = ones(100,1)*50 + rand(1,100)'*0.1;      % generate x values with noise

y = (-(1:100) + (rand(1,100)-0.5)*10)';     % generate y values with noise

scatter(x,y); axis([0 100 -100 0]);         % visualize data points with proper scale ratio (1:1)

</code></pre>


<ol class="questions">

<li>Write MATLAB/Octave code to estimate the parameters of a line that best fits the data. See slides 29&ndash;30 in Feature Matching and Robust Fitting slides (<a href="http://cse.taylor.edu/~jdenning/classes/cos351/slides/09_featurematching.html#sect029" target="_blank">link</a>).</li>

<li>Draw the line over the scatter plot using the <code class="matlab">line</code> function, and post your figure.</li>

<li>In English, explain how this data set is different from the similar problem on the first take-home exam.  How did you modify your model and code to better predict this data?</li>

</ol>

</div>


<div class="answers" id="a5">

<ol class="answers">

<li>(write your answer here)</li>

<li>(write your answer here)</li>

<li>(write your answer here)</li>

</ol>

</div>


<!--  ========================================================  -->

<!--  ========================================================  -->


<h3 id="h6">q6. Optical Flow (12pts)</h3>


<div class="questions" id="q6">

<p>Below are two frames form the Blender Foundation movie Cosmos Laundromat (<a href="https://youtu.be/Y-rmzh0PI3c?t=444" target="_blank">source</a>).</p>


<figure><img src="cosmos-laundromat-0.png" height="150"><img src="cosmos-laundromat-1.png" height="150"></figure>


<ol class="questions">

<li>Write MATLAB/Octave code to estimate the optical flow from the first image to the second. Use the <code class="matlab">quiver</code> function to draw the field (see <a href="http://www.mathworks.com/help/matlab/ref/quiver.html" target="_blank">quiver</a> for ref).  Refer to slide 37 of the Optical Flow slide set (<a href="http://cse.taylor.edu/~jdenning/classes/cos351/slides/15_opticflow.html#sect037" target="_blank">link</a>).</li>

<li>Describe what issues you ran into and how you worked around them.</li>

</ol>

</div>


<div class="answers" id="a6">

<ol class="answers">

<li><pre><code class="matlab">% write your code here</code></pre></li>

<li>(write your answer here)</li>

</div>


<!--  ========================================================  -->

<!--  ========================================================  -->


<h3 id="honor">Honor</h3>


<p>Enter your name and the date in the <code>&lt;div></code> element below to indicate that you have followed all of the rules outlined in the <a href="#instructions">Instructions section</a> above, that you have maintained academic integrity, and that the work you are turning in is your own work.</p>


<div id="signature">Elise White, 5/2/16</div>


</body>

</html>