Description
Title: CNN for Image ClassificationProject Proposal: CNN for Image Classification (Standardized Term Project)Objective: The aim of this project is to design and implement a Convolutional Neural Network (CNN) for image classification. We will use a dataset of cat and dog images and build a CNN classifier to distinguish between the two categories. Additionally, we will explore the impact of adding dropout and batch normalization layers to improve the classification performance.Project Plan:Data Collection and Preprocessing: Download the cat vs. dog image dataset containing 25,000 images (12,500 for each category) from the provided URL: http://www.cse.fau.edu/~xqzhu/courses/cap6619/dataset/train.zip.Unzip the downloaded file to access the dataset. Organize the data into training and testing sets.Model Architecture: Create a CNN classifier with the following architecture: At least three convolutional layers. Two pooling layers. Two dense layers. Use appropriate activation functions, kernel sizes, and strides for the layers. Visualize the architecture using a flowchart to illustrate the roles and relationships of each layer.Training and Evaluation: Train the CNN model using the training dataset. Evaluate the model’s performance on the test dataset and report the classification accuracy. Record the loss and accuracy during training to analyze the learning progress.Enhancing Model Performance: Extend the initial CNN model with: Dropout layers to prevent overfitting. Batch normalization layers for improved training stability. Experiment with different configurations of dropout layers (number and rates) and batch normalization parameters.Performance Analysis: Report the classification accuracies and losses for different configurations of the CNN model. Analyze the impact of convolution layers, dropout layers, and batch normalization on the model’s performance. Visualize the results in tables and plots.Deliverables: Term Project Report: A comprehensive report detailing the project, including explanations of CNN structures, the roles of convolution layers, pooling layers, dropout layers, and dense layers. A flowchart or graphical representation of the CNN architecture. Tables or plots showcasing classification accuracies and losses for various model configurations. Analysis of the results, focusing on the effects of different layers and parameters on model performance.
Unformatted Attachment Preview
CAP 6619: Deep Learning
Standardized Term Project: CNN for Image Classification
Convolutional Neural Networks for Image Classification
The goal of the standardized project for students to exercise on designing deep convolutional neural
networks for image classification task.
Students can use this task as their term project. You are also encouraged to propose and work on your
own project (please refer to “term project proposal announcement” for details). If you choose to use
Standardized Term Project, you need to follow instructions bellows:
1. Students still need to submit your term project proposal (just specify that you will use
standardized term project), otherwise you will lose one-point proposal credit.
2. Students can only work by yourself (cannot form a team).
3. Students cannot use standardized term project to substitute the final exam (which means that
you need to participate in the final exam).
4. Students still need to submit your term project report (Please refer to “Course Project
Instruction” for details).
CNN for Image Classification Term Project Instruction
1. Please refer to the CNN Lecture and Notebook for using/designing CNN for image classification.
2. Please download the cat vs. dog images (556MB zip file) from the following URL
http://www.cse.fau.edu/~xqzhu/courses/cap6619/dataset/train.zip
3.
4.
5.
6.
You can also download images from the Kaggle site (https://www.kaggle.com/c/dogs-vscats/data)
Unzip the downloaded (zip) file. There are 25,000 (dog and cat images, 12,500 for each
category) in the “train” folder.
Please follow “CNN for Image Classification [Notebook html] ” to create your own CNN image
classifiers and validate their performances.
Create a CNN classifier with at least three convolution layers, two pooling layers, and two dense
layers, and report the performance of the classifier on a test set.
For the same network created on step (5), please add dropout and batch normalization layers
(and different parameters), and validate the classification performance with respect to one set
of parameter changes (e.g., different kernel sizes, different number of dropout layers, or
different dropout rates).
Term Project Report Instruction
1. You will need to follow the “Course Project Instruction” to prepare your term project report.
2. Your report should include following major components:
a. Explain CNN structures, the roles of convolution layers, pooling layers, dropout layers,
and dense layers.
b. Draw a graph (or a flowchart, or figure) to show roles and relationships of each layer.
c. Report classification accuracies (or loss) of your CNN classifiers as tables or plots, and
analyze the performance of the classifiers with respect to the selected convolution
layers and dropout layers.
CAP 6619: Deep Learning (2023 Fall)
Course Report Instruction
(For students participating in final exam) (10 points)
Due date [December 10 2023, Firm]
This instruction only applies to students who will participate in the final exam (i.e., do NOT
wish to substitute the final exam using term project). If you wish to substitute the final exam
using a longer version of the term project report (i.e., a research report), please follow
“Research Report Instruction” (for students substituting final exam) in the Canvas.
The grading of the term project report is based on the following criteria.
1. Overall [2 pts]: You MUST organize your report in IEEE format, with 1000 words minimum. Please
note that table/figure do not count towards the word limitation. You can use IEEE word or Latex
temperate from the following URL
a. Template: http://www.ieee.org/conferences_events/conferences/publishing/templates.html.
b. Plagiarism: You cannot copy any sentences, paragraphs, or figures, from any external
sources (such as published papers or Internet). If Turnitin indicates that a submission is over
30% similar to any other submissions, the report will receive a penalty score calculated using
formula listed in the term project announcement.
c. If you have to cite a figure/graph published somewhere else, please properly cite the source of
the reference [0 credit if plagiarism check returns over 50% similarity to any published work].
[Grading of grammars and typos are included in the “Overall”]
2. Introduction [1 pt]: Your report should have an introduction section with about 200 words. The
introduction should clearly state (1) what is the research problem to be studied in the report; (2) the
motivation of the problem studied in your report; (3) how are the problem solved by existing
methods, if any; and (4) a brief description about the method you will propose in the report. Cite at
least 5 references related to the project in the Introduction [200 words]
3. Main body [3 pts]: In the body of your project, you will need to provide technical details of your
design [500 words]
a. If your report is about new design to solve a research problem, you will need to describe
motivation of your designs, and detailed framework. Use flowcharts, figures, or some pseudocode to describe your algorithm details. [Use at least one figure (or flowchart) to demonstrate
the system framework or architecture]
b. If you report is using standardized term project (CNN for image classification), this section
should report aspects related to convolutional filters, the architecture of your CNN networks,
any changes or new designs in your framework. [Use at least one figure (or flowchart) to
demonstrate the system framework or architecture]
c. If your report is about experimental studies, you will need to provide a brief description about
your learning/classification methods, the benchmark datasets, and different measures applied.
You should also explain how the experiments are carried out in your study, and what type of
empirical study goals you intend to achieve.
4. Experiments [2 pts]: In the experiments, you need to introduce (1) main purpose of the experimental
studies; (2) what are the tools used to design the algorithms; (3) what are the baseline methods for
comparisons; and (4) what are the performance measures and data used for empirical studies. You
should also use figures and tables to report the results collected from your studies, and summarize the
experimental results [200 words].
a. Experimental settings: including programming tools/languages, the setting of the parameters
used for different methods. [1 pt]
b. The results: The detailed results reported in figures/tables with necessary analysis and
descriptions. You will need to include at least one figure and one table to show the results. [1
pts]
5. Conclusions [1 pt]: In the conclusions, you should briefly summarize the research problem studied in
your report. Explain what you have done and summarize the major findings. Draw any informative
conclusions, which can be useful to guide the followers [100 words, 1 point]
6. References [1 pt]: Please cite at least 5 relevant references in your report. At least 3 of them must be
from 2017 or after.
Purchase answer to see full
attachment
