LAB: CNN Object Detection 2

LAB: CNN Object Detection (Free Topic)

I. Introduction

In this lab, you are required to create a simple application that uses a CNN-based object detection. You are free to choose the application topic of your interest.

Application Examples

Choose a topic relevant to Mechatronics. Refer to past lab reports on object detection

YOU MUST NOT choose the same topic of previous labs

• Kaggle Data Challenge

• Defect detection

• Face recognition

• Pedestrian/Vehicle/Signpost detection for autonomous driving

• People/vehicle Counting

• Anomaly detection and more

Guidelines

The whole code should be programmed using OpenCV-Python and Pytorch.

• DO NOT copy a project from online sites.

• You can refer to any online material and github repository for assistance and getting ideas with proper reference citation.

• You can use any pretrained object detection model, such as YOLO v3~v5, etc..

• You may also train the model using custom datasets

• You can clone a github repository of the object detection model(e.g. YOLOv5), as long as you cite it in the reference.

If you create a simple hardware for demonstration, you will get extra score. Please ask TA if you need an embedded GPU board or any other hardware materials.

II. Report and Demo Video

This lab will be scored depending on the Contents, Complexity, and Completeness .

You are required to write a concise report and submit the program files and the demo video.

Report

The lab report must be written as a 'Tutorial' format to explain the whole process A to Z in detail. Your report will be posted in the class website and be open to public.

A high score will be given if a reader should be able to follow the report instructions and get the same results.

Use the report template given here: https://ykkim.gitbook.io/dlip/dlip-project/report-template

• Also, see example tutorials: example 1, example 2

Requirement

• Write the report in markdown ‘*.md’ format

• You need to include concise explanations and codes for each process in the report

• Create a Repository(Public) in Your Github

  • Repository example: DLIP_Project_keywords_2022

  • e.g. DLIP_Project_MaskDetection_2022

• Upload the source code and report in your github repository

• Then, download your repository as zip file and submit the zip file to online.handong.edu. The Zip file includes

  • Report (*.md)

  • Report (*.pdf)

  • src (source code)

  • images (sample, test image files )

  • data (whole dataset or download link)

Demo Video

You must create a demo video (~40 sec) that will be uploaded in the Lecture's Youtube channel.

• The demo video must contain

  • Title page: course name (DLIP2022-1 by Y.-K.Kim), your name, your project tile, date

  • Introduction, Dataset, Model, Results, Conclusion

• Embed the youtube video in your report

• Submit the video file to LMS or TA's email