Projects are a fundamental building block of the platform. Think of Projects as a sort of folder for collecting all the things you need to create and manage a single computer vision model.

For our platform, a Project is a discrete unit of work that encompasses all the things you need to create a new computer vision model or to update an existing one.

To do this, a Project associates all the input data you need (media from your Datasets) and, based on a few settings you select when you create the Project, sets you up to quickly and easily create the training data you need to build or update a model.

Following the baking analogy I've used in previous articles: think of a Project as a bakery devoted to one specific baked good. You'll work with your suppliers (Datasets) to get your main ingredients, then prepare those ingredients (training data) and use them to bake something (train a model). Once your goods are baked, you can inspect them for quality (testing & evaluation), then put the best ones on display for your customers to enjoy (productionize).

Because a Project is devoted to one specific computer vision model, it's best to keep your Projects very well scoped and defined. Just like a human, the model that comes out of your Project will perform better if it's been trained to look for one or a few similar objects on similar media.

So don't create a "Detect Everything" Project: go instead for a "Bottle Cap Defect Detection" Project, for example. After all, you can create as many Projects as you want, so there's no need to force mismatched use-cases together.