[Week 1- Origin of Food]

The Basic Idea

Considering the wide range of information we have reached from numerous external sources, we have roughly devised the possible paths we are going to follow for our course project and started to ponder and focus on the accurate approaches to work on. The coarse idea behind our project is devising an algorithm that has the capability of guessing the origin of the food that was given as a set of input images.

Computer Vision

After coming up with this idea, It was indisputable that computer vision approaches were going to be an integral part of our project. As we have seen the speech of the director of Stanford’s Artificial Intelligence and Vision Lab, Professor Fei-Fei Li, during her talk she has emphasized the vital importance of the idea of using the huge amount of data as training data since there is not a constant mathematical structure that we can give the computer to follow in order to understand pictures as we do. She also mentioned that this was one of the leading approaches that came up with the idea of using Big Data as training data and how this huge amount of data is collected by establishing ImageNet. Afterwards, she explains their machine learning algorithm and how the convolutional neural network blossomed and suprisingly became the winning architecture to generate accurate results in object recognition. Her way of explaining these steps in a quite simple way enabled us to have a basic understanding of our problem and showed us the methods to focus on.

After examining this video that explains the basic level definition of the problem and solution, and our small talk with our advisor Aykut Erdem , we had a more concrete path to follow and found out that we have to go through deep learning methods.

First Step

The first problem we encountered was that we had a set of images but did not have tags which defines the name of that particular country or region. So we have to label these images on our own. The first step to take was placing these labels by hand or trying to find a way to automate this process to simplify the operation which we will talk about in our next progress reports.