Week 3— DeepNutrition: Be Aware Of Your Macronutrients
Hi,
In this week I want to talk about our dataset.
We are decided to use the Food-101 Dataset which is provided by Lukas Bossard, Matthieu Guillaumin, Luc Van Gool from ETH Zurich. The challenging dataset consists of 101 food categories, with 101'000 images. For each class, 250 manually reviewed test images are provided as well as 750 training images. The training images are not cleaned, and thus still contain some amount of noise. This comes mostly in the form of intense colors and sometimes wrong labels. All images have a maximum side length of 512 pixels. The total of the dataset is 5GB. But we are thinking to train a model for detecting aboutt ~15 classes which makes ~742 Mb.
We also continue our research process by reading articles.It is obvious to use CNN is a good idea in our current problem.But our main goal is to determine the top N CNN models and build them in our machine and test with our data. We will discuss our results of paper reading in the next weeks blogpost.
