Gamification of Annotations

Data curation and annotation is a component of machine learning and computer vision projects. This project explores the application of games to the data annotation process.

This solution addresses the need for scale in data collection. The goal is natural adoption by gamers by building sticky and challenging games.

In this example, the data is being annotated to train a color similarity model. Fashion images are uploaded into the system and the individual images are broken up into patches. These patches are compared to each other through the gaming component. Visually similar patches will get connected to each other by the drawn lines in the game.

These patches are sorted into high level color clusters and formed into a game. Each game is presented to multiple gamers where they can connect the dots to collect points. They get higher points for connecting dots which agree with other annotators. Annotations from multiple gamers are aggregated to determine which dots are visually similar to each other. Apparel with high number of connected dots are considered to be similar.