[WEEK-1 — Clean/Messy Rooms Detection]
Theme: Scene recognition and Room classification
Team Members: Damla Ünal, Zekeriya Onur Yakışkan, Atakan Erdoğdu,Didem Yanıktepe
Introduction: Scene recognition is one of the determining tasks of the computer vision. Our project focuses on this task. In this project our aim is to classify parts of a house(bedroom, kitchen etc.). In addition to that, we will try to determine whether room is messy or clean.
Considering a hotel, there are many rooms and customers generally wants to find their rooms clean. This task involves finding messy rooms in the hotel. Our project, may help this issue and automatically find the rooms that needed to be cleaned.
Each rooms need different cleaning products. Another application of this project might be help advising cleaning products to customers with respect to which rooms are messy in their house. This might be very helpful with in the context of smart homes.
Datasets:
In this project, we will create our own dataset. we have photos from various sites for clean rooms. We will combine these photos with messy room photos that we will collect from Google and create our own dataset.
Here is the list of datasets where we collected our data: ADE20K, Image-net.
We have six classes. These classes are messy bedrooms, messy kitchens, messy living room, clean bedroom, clean kitchen, and clean living room. We target a dataset consists of at least 1000 pictures for each class.
Approach:
We will try two diffrent aproaches and combine them later. The first approach is convolutional neural networks. Our other approach is support vector machines. Then we will try to combine them.We will compare results of this approaches and try to conclude which one is better at this task.
Related Works:
