[Week4 - Where is this, in Ankara?]

Team Members: Alper Adil Çetinkaya, Esme Şule Alp

Following the lack of progress last week, this week, we’ve steped up our game, starting our data collection but not before finally reached some conclusions about our project.

As a method, we’re going to use deep learning. Because CNNs have high success rate for image geolocalization and there are previous works which we can benefit from. Here is the paper we want to mainly follow:

PlaNet — Photo Geolocation with Convolutional Neural Networks

We’ve also reached a conclusion for our dataset. We’re going to create a dataset which consists of well known places of Ankara. At first, we had planned it as a more general Ankara dataset however due to time restrictions, we have to keep it as well known places. But we always have a chance to enrich our dataset later on. The other change in our plan is that we thought we could get our dataset from Google Street View; but it seems we’re going to use it only for exceptions. Instead we will search for our data on our own through Google’s image search as we felt it made things easier and more fun.

We had to make a decision about how our dataset’s format should be, before starting to collect. We’ve looked for mostly used image datasets to understand their format, and here are some examples from our data, hopefully we’ve decided the suitable format:

So as you can see this week was very productive for our project, we hope to continue and improve upon this productivity in the future. Our current goal is to finish collecting our data by this Wednesday, so we will have much to say about that next week. Thanks for sharing this journey with us, hope to see you again next week.