There are various photographer marketplaces on the web, which help professionals find interesting gigs that match their skills. These can be photoshoots for a new restaurant's dishes, a wedding, or even a new phone that's coming out. In most of these cases, the pictures are meant for private or commercial use, but I had never heard of images being captured explicitly for a machine's enjoyment, until now. Microsoft’s Trove project helps connect developers with people, who could supply photos to train machine learning models.
In order for a machine to "learn," it needs a considerable amount of data, such as photos. In optimal conditions, these need to be diverse, of high quality, and sometimes match specific requirements, which is why stock images aren't always a good option. For instance, most publicly available photos of dumbbells would show them being held by someone, making AI assume arms are actually part of dumbbells. To avoid this, developers need a bunch of new shots and rely on a high number of people to provide them. Unfortunately, it's not easy to consolidate all these images, and in many cases, they end up not matching specific requirements that are essential to the developers.
To avoid these issues, Trove's marketplace aims to connect contributors with AI developers and help them communicate effectively and understand the context before submitting images. People therefore get the opportunity to contribute to an AI project by submitting pictures from their camera roll or shooting new ones. Most importantly, contributors get paid for their photos and can easily receive their funds through their PayPal accounts.
When it comes to developers, this gives them the opportunity to crowdsource specific content for their project, but also create enthusiasm around their work. They can also provide feedback to contributors to make sure the images match their criteria.
Microsoft ensures it provides a license that will protect the ownership of photos uploaded onto the platform, as well as transparent information about how data will be used.
Trove is currently accepting a small number of participants in the United States on both Android and iOS. Users based in the United States can request an invite to join the experiment as a contributor, or request to add a machine learning project to the experiment.