[Week-2 Clean/Messy Rooms]
Team Members : Didem Yanıktepe , Damla Ünal , Atakan Erdoğdu , Zekeriya Onur Yakışkan
In this week, we merged data from respective datasets. For clean room classes, we used datasets Image-Net and Ade20k. However, we could not find any dataset for messy room classes. Therefore, we wrote a script for collecting images from google. We used the selenium web driver in this script. Selenium is a library for web crawling. This script takes two inputs, a number, and a query. It searches the given keyword in google images and downloads that number of images.
Ade20k separates its dataset as train and validation part. In Ade20k dataset, we have 652 images of clean kitchen, 697 images of clean living room and 1389 images of clean bedroom for the training phase. In the validation part, we have 338 clean kitchen images, 320 clean living room images, and 628 clean bedroom images.For messy images we crawled it from google.
Resources:
http://groups.csail.mit.edu/vision/datasets/ADE20K/
http://www.image-net.org/synset?wnid=n04351550
