Scribd
Upload
52 views2 pages

Computer Vision (600.461/600.661) Homework 6: Segmentation and Recognition

This document provides instructions for Computer Vision homework 6 at Johns Hopkins University. It consists of two main parts: image segmentation and object recognition. For image segmentation, students are asked to implement spectral clustering and use it along with k-means clustering to segment images based on intensity and color. For object recognition, students are instructed to use code from tutorials to perform instance recognition of the Eiffel Tower from images and category recognition of airplanes and motorbikes from datasets. Students are told to submit their code, answers, and instructions in a zip file with a specific directory structure by email.

Uploaded by

Mohiuddin Ahmad
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
52 views2 pages

Computer Vision (600.461/600.661) Homework 6: Segmentation and Recognition

This document provides instructions for Computer Vision homework 6 at Johns Hopkins University. It consists of two main parts: image segmentation and object recognition. For image segmentation, students are asked to implement spectral clustering and use it along with k-means clustering to segment images based on intensity and color. For object recognition, students are instructed to use code from tutorials to perform instance recognition of the Eiffel Tower from images and category recognition of airplanes and motorbikes from datasets. Students are told to submit their code, answers, and instructions in a zip file with a specific directory structure by email.

Uploaded by

Mohiuddin Ahmad
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Computer Vision (600.461/600.

661)
Homework 6: Segmentation and Recognition
Instructor: René Vidal
Due 12/02/2014, 11.59PM Eastern

1. (25 points) Image Segmentation

(a) (10 points) Implement the Spectral Clustering algorithm as follows:


Function [group, evals, evects] = spectral(W,K,method)
Parameters
W N × N similarity matrix
K the number of groups
method the method to be used from ‘unnormalized’, ‘normalized’ and ‘symmetric’
Returned values
group 1 × N vector with group membership of each point
evals 1 × K vector of the smallest K eigenvalues of the graph Laplacian
evects N × K matrix of the corresponding eigenvectors of the graph Laplacian.
Description
Computes the clustering of points using spectral clustering algorithm.
‘unnormalized’ uses the graph Laplacian, L = D − W
‘normalized’ uses the graph Laplacian, L = I − D−1 W
‘symmetric’ uses the graph Laplacian, L = I P − D−1/2 W D−1/2
N
where D is the diagonal degree matrix with di = j=1 Wij .
You may want to code it in such a way it exploits the sparsity of W whenever the matrix W is sparse.
(b) (7.5 points) Intensity-based. Use the kmeans and spectral (symmetric) methods to segment the
images on the course webpage using intensity only. For kmeans, use the building matlab function with
multiple restarts. For the spectral use your own code with a k-connected neighborhood so that W is
sparse. Feel free to choose appropriate values for σI and k in defining similarities for intensity and number
of neighbors, respectively. Comment on the performance of the different methods on the different images.
(c) (7.5 points) Color-based. Use the kmeans and spectral (symmetric) methods to segment the images
on the course webpage using color only. Use both the (r, g) and the (H, S) color representations. Feel
free to choose appropriate values for σc and k in defining similarities for color and number of neighbors,
respectively. Comment on the performance of the different methods on the different images.

2. (25 points) Object Recognition


(a) (10 points) Recognition of Object Instances. Go to images.google.com. Type “tour eiffel”. Download
the top 11 images. Using image 11 as a template, your goal is to match (recognize) the template to each one
of the first 10 pictures. For this purpose, please read and follow all the steps in Vedaldi’s object instance
recognition tutorial. You should submit a script called hw6q2a.m that loads the images, calls Vedaldi’s
code, and plots the results. If your script calls a new function that you have created, then you should also
submit this new function.
(b) (15 points) Recognition of Object Categories. Given images of airplanes and motorbikes, your goal is
to train classifiers for determining whether a novel image contains an airplane (motorbike) or not. For this
purpose, please read and follow all the steps in Vedaldi’s object category recognition tutorial. You should
submit a script hw6q2b.m that loads the images, calls Vedaldi’s code, and gives the results. If your script
calls a new function that you have created, then you should also submit this new function. For each dataset,
please use the first 200 hundred images for training, and the remaining 600 for testing.

1
Submission instructions. Send email to [email protected] with subject 600.461/600.661:HW6 and attachment
firstname-lastname-hw6-vision14.zip or firstname-lastname-hw6-vision14.tar.gz. The
attachment should have the following content:
1. A file called hw5.pdf containing your answers to each one of the analytical questions. If at all possible, you
should generate this file using the latex template hw1-vision14.tex. If not possible, you may use another editor,
or scan your handwritten solutions. But note that you must submit a single PDF file with all your answers.
2. For coding questions, submit a file called README, which contains instructions on how to run your code. Use
separate directories for each coding problem. Each directory should contain all the functions and scripts you are
asked to write in separate files. For example, for HW2 the structure of what you should submit could look like

(a) README
(b) hw2.pdf
(c) hw2q3: hw2q3c.m, hw2q3e.m
(d) hw2q4: hw2q4b.m, hw2q4c.m
The TA will run your scripts to generate the results. Thus, your script should include all needed plotting com-
mands so that figures pop up automatically. Please make sure that the figure numbers match those you describe
in hw2.pdf. You do not need to submit input or output images. The output images should be automatically
generated by your scripts so that the TA can see the results by just running the scripts. In writing your code, you
should assume that the TA will place the input images in the directory that is relevant to the question solved by
your script. Also, make sure to comment your code properly.

You might also like