search

COSC481W-2020Fall / cosc481w-581-2020-fall-convoluted-classifiers

cosc481w-581-2020-fall-convoluted-classifiers created by GitHub Classroom
MIT License
0 stars 1 forks source link

readme

Dog Breed Detector v0.1.0

This is an open source dog breed detector.

Dog Breed Detector App Quick Access:

On your Android Device, please click here to download the file: app-debug.apk

Once the app-debug.apk file is downloaded you will be prompted to open the file and install the application.

** Note: You will need to enable the ability to install APK's from "unknown sources" (Typically found in Settings > Applications > "Unknown Sources" checkbox) or you may be asked while downloading the application.

How to install Android Studio:

Download Android Studio https://developer.android.com/studio

Run the Android Studio Setup:

  1. Under “Choose Components”, leave as default and continue
  2. Choose the install location on your computer and continue
  3. Under “Choose Start Menu Folder”, leave as default and continue
  4. Wait for installation to complete and click Next
  5. Launch Android Studio
    • When prompted to import Android Studio Settings select “Do not import settings”

Now we will go through the Setup Wizard:

  1. Select “Standard” for the Install Type and click Next
  2. Select the “UI Theme” with your preference and click Next
  3. Click Finish

How to load the CNN Project from Github:

  1. Open Android Studio
  2. Choose “Project From Version Control”
  3. Enter the Repository URL:
  4. Click Clone
  5. Select "Create project from existing sources" and click Next
  6. Enter a project name or leave as default and click Next
  7. Click Finish
  8. It will ask if you want to open the project, click Yes
    • It will take a few minutes for the Gradle to sync

How to run app on your own device through Android Studio:

  1. Connect your device to your computer using a USB cable

On your device:

  1. On your device open up “Settings”
  2. In Settings go to “About Phone”
    • Certain versions will have it under System > About Phone
  3. Scroll to the bottom and tap on “Build Number” 7 times
  4. Return to Settings and go System > Advanced
  5. Open “Developer Options”
  6. Turn on “USB Debugging”
    • It will ask to “Allow USB Debugging”, Click Allow

In Android Studio:

  1. In the toolbar, select the device that you want your app to run on from the “Target Device” drop-down menu
  2. Click Run to the right of the “Target Device” menu

Install Prototype

  1. Install python3.7 from python.org
  2. Clone/download the repository
    • Download zip
    • git clone https://github.com/COSC481W-2020Fall/cosc481w-581-2020-fall-convoluted-classifiers.git
  3. Create a venv (using bash) (Windows recommended using git terminal, or ubuntu shell using WSL)
    1. python3.7 -m venv venv
      1. if previous does not work on windows, use py -3.7 -m venv venv
    2. Mac/Linux - venv/bin/python -m pip install -r requirements.txt
    3. Windows - venv\Scripts\python.exe -m pip install -r requirements.txt

Run Prototype

  1. Run command venv\Scripts\python.exe main.py --test_location="[LOCATION]"

Prototype Description

For our prototype, we would like to create a way to automatically detect a canine animal inside an image file using Python. We will have a folder in our Github repository containing approximately 100 canine images that our program will search through. We chose to start with 100 images so that we had a large enough sample size to test with. Our Python program will use the tensorflow, openCV, keras, and imageAI modules and it will be able to detect which of the images in that folder are of canines and output "yes" if it is a dog and "no" if it is not.

The user will simply be prompted to enter the given file/directory that they wish to examine. After the file/directory is entered no more prompts are given and no more input is needed from the user.Using the os module in Python, the program will be able to accept both image files and directories. If the user selects for a file, the program will output just a "yes" or "no". If the user selects for a directory, the program will output the image filename followed by a "yes" or "no". The program will also create and output files in the hardcoded directory ./.output/dog{n}.png, where n is the number of images found with dogs, that will contain the images entered by the user now each marked with a box around the dog that was detected in every positive image file. The program will then exit. If no valid image extensions are found, the program will say "no valid images" and will exit.

We will be using the command line prompt to prompt the user, provide user input,run the program, and display any output. To verify that our program is accurate we will manually check the output against the correctly labeled images that do identify a dog.

Prototype use case diagram