[WEEK 2- Predicting the pieces of furniture in living rooms]
Group Members: Mohammed ALI, Aybüke Yalçıner, Hatice ACAR.
As we mentioned week 1, “ our goal in this project is to predict the pieces of furniture in a living room”
Dataset
This week, we have collected our dataset. The dataset is formed the top 10 furniture images — sofa, armchair, chair, coffee tablet, lamb, television, television table, carpet and pictures/paintings — in living rooms. For each category, we collected the minimum 300 images. All images are separated in folders.
Preprocessing the Dataset
We collected all images from different sources so images are in different sizes. To make them in the same size we made some search and we found that python can do it easily for us.
The python has 2 different libraries to do it OpenCV and PIL. Using these libraries we first tried to crop the images. We firstly worked on only one image. And we cropped it successfully. Then we tried to do it for all photos in a for loop. In the early times it cropped the images one by one but after some time we got a bad error because of there are too many images to crop and after cropping, computer opened the cropping image on the screen automatically so we tried to resize images instead of cropping.
It is worked successfully and now our all images have the same size which is 200x200. We determined this size because the minimum size of the original images is 225x225. In addition, to get more accuracy at the end of this project, we convert all images in the same extension, which is jpg, using the PIL library.
Algorithm
To get started the project, we have done some research that is about which algorithm is the best to classify the images. And we encounter with some Machine Learning algorithms line k-NN, SVM, Logistic Regression etc. When we compare these algorithms with CNN(Convolutional Neural Network), we see that CNN is better because of there is no transfer learning does happen in Machine Learning algorithm, Machine Learning algorithms’ learning is less while comparing to CNN. In CNN, the learning is more and also less error is occurred so we will use the CNN algorithm while this project.
Choosing the right framework
After we have done some research, we have decided to use “Keras”. Keras is a high-level neural networks API, that allows for easy and fast prototyping and supports both convolutional neural networks and recurrent networks, as well as combinations of the two. we started learning how to use it and implement it to get the job done.
