PowL-Pets-own-bowL

Graduation project

Notice

At first, we were tried to make it with raspberry pi.
But we couldn't do it for some reason.
The python program will replace the raspberry pi to show flow of our project.
We hope that you will take that into consideration.
Thanks

---

You have to download python code at here
save them at the same directory of this repository

---

Python

You have to change your flow.py 's IP address code to your current pc's IP address
flow.py Line 12, HOST ="your PC ip address"

---

Android

Also, you have to change your com.powl.graduation.ui.gallery.java and com.powl.graduation.enroll.java's IP address code to your current pc's IP address
com.powl.graduation.ui.gallery.java Line 64, String ip ="your PC ip address"; com.powl.graduation.enroll.java Line 22, String ip="your PC ip address";

---

Presentation video

Youtube_video

Requirements

tensorflow = 1.13.1
opencv-python = 4.1.2
PIL = 7.1.2
keras = 2.2.4
tqdm = 4.46.1
beautifulsoup4 = 4.9.1
selenium = 3.141.0
tensorboard

Datasets

Basic datasets are from here
And Added some crawled data
There are 16 kinds of species
And 6160 pictures in total

Getting started

Installation

Clone the git repository.
We will call the cloned repository/python/Tensorflow-Dog-Breed-Classifier/ as $home

git clone https://github.com/seokhyeonSong/PowL-Pets_own_bowL.git
cd PowL-Pets_own_bowL.git

Crawling (optional)

This can be skipped
To make dataset more rich, you can crawl some data from google by this code
Crawled data will be at /dog_images/each_folder
You have to distinguish some garbage images from dataset
num_images are used to set the number of crawling image
default is 50

python crawling.py [--num_images number_of_images]

Retraining (optional)

This can be skipped
To retrain the features of dogs, you can use this code

python retraining.py --image_dir /dog_images

This code is came from here

Run

You have to download our application to register
Or you can just enter your dog's breed as insert value with args

Dog's breed	insert value	Dog's breed	insert value
Beagle	beagle	Border Collie	border collie
Chihuahua	chihuahua	Cocker Spaniel	cocker spaniel
Doberman	doberman	French Bulldog	french bulldog
Golden Retriever	golden retriever	Labrador Retriever	labrador retriever
Maltese	maltese dog	Papillon	papillon
Pekinese	pekinese	Pomeranian	pomeranian
Schnauzer	schnauzer	Shih Tzu	shih tzu
Poddles	standard poodle	Yorkshire Terrier	yourkshire terrier

python flow.py [--breed breed_insert_value]

You can register your dog with application
Your phone and notebook should be in same Wi-Fi

To register your dog

1. run flow.py and select server on mode
2. send your dog's picture by album or take a picutre
3. You will receive your dog's breed prediction from server you can choose your dog's breed
4. Then your registered dog will be added to breed.txt

To run PowL

1. Set your labtop front of dog's bowl Before you run PowL mode, the camera's scene should be static
2. Run PowL
3. Then program will run, and check whether you've selected breed it or not

Evaluate

notice

This codes are from here

Run

You can see the training and validation result with tensorboard

1. Enter here
2. Run this code for train result : tensorboard --log_dir=$home/../tmp/retrain_logs/train
   for validation result : tensorboard --log_dir=$home/../tmp/retrain_logs/validation
3. You can see the result at here

Citation

author : @AthulDilip, @sabeersulaiman
title : Tensorflow Dog Breed Classifier
publisher : {GitHub}
journal : {GitHub repository}
howpublished : https://github.com/AthulDilip/Tensorflow-Dog-Breed-Classifier#tensorflow-dog-breed-classifier