Why this dataset?
Image recognition software can identify what’s in a picture, such as food or clothes.
Unfortunately, this software often gets confused by the visual differences between countries and cultures.
The Everywhere Dataset is an open collection of pictures from people all over the world to help this software recognize the differences.
Who made this?
Hi, I’m Niels. I traveled quite a bit around the world and met so many nice people in different geographies and cultures.
This open dataset is my attempt to give back to them, as I received so much from them. You can read more about me on my personal website.
The Everywhere Dataset is an open validation image dataset containing pictures from (hopefully soon) every country in the world. As of 23-03-2021, it contains pictures from 11 countries, with 37 images of 24 assets in 1 category (food), 67 labels and 147 tags.
*This is a VALIDATION dataset, by downloading you agree to only use it for validating models and NOT using it for training purposes.
How is the data collected?
All data is collected ‘by hand’. This makes for a very intense collection process, but ensures the data is original, has context and includes the right labels.
In particular, I am having a ‘Digital Meal’ with a person in every country of the world. So, basically, video chatting while sharing breakfast, lunch or dinner and then asking the person what they are eating.
Right now, I ask friends of mine to have a digital meal with me. Soon, I’ll have to start asking friends of friends. Of course, this sampling procedure brings biases with it, which will be described in detail as the project progresses.