dijitalYoruk/Indoor-Transfer-vs-Scratch-VGG dijitalYoruk/Indoor-Transfer-vs-Scratch-VGG

  • Stars
    star
    3
  • Rank 3,963,521 (Top 79 %)
  • Language
    Jupyter Notebook
  • Created over 5 years ago
  • Updated about 5 years ago
Twitter logo Share LinkedIn logo Share Facebook logo Share
dijitalYoruk/Indoor-Transfer-vs-Scratch-VGG
github

dijitalYoruk

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Indoor scene recognition is a challenging open problem in high level vision. In this project, two VGG16 models sre used to classify the indoor dataset. One of these models is pretrained and the other is not pretrained. It can be obviously seen that the pretrained model performs much better than the untrained one. The reason behind is that, when a model is pretrained, learning new tasks is easier for it. The pretrained model had an acurracy of 86% on test data with 10 epochs, whereas, the model that is not pretrained had 40% acurracy on test data even though it had 80 epochs.

More Repositories

1

InstaKotlinApp

InstaKotlin is a social media application that enables people to share their images and videos. Users can chat with each other and make comments about the posts of other users. Used Kotlin, Fragments, Material Design, Firebase, various third-party libraries, Git, consists of 15k lines of code.
Kotlin
9
star
2

Caltech-UCSD-Birds-PyTorch

German traffic signs dataset is analysed with Keras. This task is a multi-class classification problem and the dataset is complex because of containing 200 classes with 12000 images. The dataset is splitted into training, validation and test dataset and normalised. Data is preprocessed and data augmentation is applied. The ResNet model is used by the courtesy of transfer learning and trained. At the end of this process, 69% acurracy is obtained on the test dataset
Jupyter Notebook
5
star
3

Catan-Desktop

The Settlers of the Catan is a strategy game played by 3-4 people where everyone plays in turns. The game starts by assembling hexagonal map pieces and shuffling and placing the cards. Each player chooses his/her figure to play with. Construction price cards are distributed to players where the information about construction such as village or cities and the points they provide are written. At the beginning of the first turn, the player who has the turn rolls two die and place villages and roads on the map to a place he desires. After each player settles his/her first villages and roads, the competition starts. In each tour, the player rolls the die and can collect sources, buy or use development cards, trade (swap sources), and do construction. After the die are rolled, players who have own constructions lying on that numbered map gains sources. There are five types of source cards which are wood, brick, iron, sheep, and wheat. These source cards are utilized while doing construction and buying development cards. If the sum of numbers on the die is seven, each player who has seven or more cards places half of them to the chest, also the player who has rolled the die can place the thief figure on wherever place on map he/she wishes where the next player won't be able to get source cards from that place. The player who places the thief gets a random source card from each player who owns construction in that place. The player who owns the tour can trade his/her source cards with other players or the chest. In order to construct a village, the player must reach that part of the game by building roads. Villages can be upgraded to cities. There are three types of development cards that add more game functionality to the game. If a player collects 10 points from constructions and cards, the player wins and the game ends.
Java
5
star
4

Classification-of-Breast-Cancer-Cells

In this project, various concepts are used for evaluating the cells and increasing the accuracy of cancer detection. These concepts are Principal Component Analysis (PCA), Logistic Regression Classification, K-Nearest Neighbors (KNN) Classification, Support Vector Machine (SVM) Classification, Decision Tree Classification, ANN and Naïve Bayes Classification. This algorithms are implemented by using scikit learn library and Pytorch.
Jupyter Notebook
3
star
5

Blogger-App-Frontend

Blogger is a social media application that enables users to share their images and blog posts. This project is the frontend part and developed with VueJS, Vuex, VeeValidate, VueRouter, Toastr, Bootstrap and axios.
Vue
2
star
6

ecommerce-microservice-app

TypeScript
2
star
7

Chatter

Chatter is a mobile application where users have the opportunity to chat with each other and also, create groups. In this mobile application Firebase Realdatabase, Storage and Cloud Messaging is used. In addition to firebase, various third party libraries and material design features are also included.
Java
2
star
8

German-Traffic-Signs-Keras

German traffic signs dataset is analysed with Keras. This task is a multi-class classification problem and the dataset is complex because of containing 40 classes with 50000 images. The dataset is splitted into training, validation and test dataset and normalised. Data is preprocessed and data augmentation is applied. The LeNet model is implemented from scratch and trained. At the end of this process, 98% acurracy is obtained on the test dataset and corresponding plots are shown.
Jupyter Notebook
2
star
9

Computer-Networks

CS 421 Computer Networks course Homeworks, Programming Assignments and sample Quiz questions.
Java
1
star
10

CS319_Lab_Assignment

Java
1
star
11

Blogger-App-Backend

An API is written for the blogger application. Implemented an OATH2 authentication system using passport. Includes policies, database relationships, migrations, validations... While implementing this application, PHP, Laravel, MySQL, Git and Laragon is used.This API is used by the fortend frameworks VueJS.
PHP
1
star
12

ALM-Todo-App

TypeScript
1
star
13

Lara-Tech

Lara-Tech is a broad technological e-commerce web site project. Two authentication systems are available for admins and clients. 2 factor authentication is also implemented from scratch. In this project, clients, admins, products, main categories, sub categories, orders, baskets, profiles, carts, comments and admin panels are available. While implementing this application, PHP, Laravel, Blade, MySQL, Git and Laragon is used.
PHP
1
star
14

style-transfer-learning-pytorch

Style Transfer Learning refers to a class of software algorithms that manipulate digital images to adopt the appearance or visual style of another image. Content image and style image are taken and resized to equal shapes. Corresponding convolutional layers of vgg19 is chosen for content extraction and style extraction and weighted with parameters. For style extraction, gram matrix of the style images are taken. Then, the loss functions for style and content images are arranged and combined. After combining, total variation loss is obtained and by backward and forward propagation, total variation loss is minimised
Jupyter Notebook
1
star
15

Psyclog

Psyclog is an application that is designed for people with mental problems. It is planned to offer a safe environment for people with issues who are afraid or ashamed of going to therapy with the aim of resolving these issues.There is a significant number of people who are timid about the idea of going to therapy because of the fear of being made fun of and the prejudice they might face. Psyclog completely eliminates these fears by providing a remote system that is usable anywhere as long as the user carries their Android device with them. If you would like to get a deeper understanding of the implementation, please check the reports.
Dart
1
star
16

CNN-on-CIFAR10

CIFAR10 dataset is analysed with PyTorch. Implemented LeNet model. This task is a multi-class classification problem. The dataset is splitted into training and test dataset and it is normalised.Data is preprocessed and data augmentation is applied. The LeNet model is implemented from scratch and trained. At the end of this process, 75% acurracy is obtained on the test dataset and corresponding plots are shown.
Jupyter Notebook
1
star