To organize data for labeling and neural net training is not an easy task. If there is no structure, things get messy 💩 pretty quick. Learn those essentials to organize your efficiently. ☝️

🗂️ Projects

Project is a combination of datasets and related meta information (like classes and tags) and it's a major building block of data organization in Supervisely.

📂 Datasets

This is where your labeled and unlabeled images, videos and point clouds live. A dataset is some sort of data folder with stuff to annotate.

🎨 Classes

Classes are pre-defined types of your annotations, for example Person or Background. Thus, every label you create has defined class.

🏷️ Tags

To associate some extra information with annotations (or images, or images, ...) you can define a Tag, for example needs_review.

Example

For instance, in Team "Driving Division" you can have a Workspace "Pre-Labeling". In this Workspace you can have a Project "Cityscapes" with two Datasets: "Zurich" and "Stuttgart".

There are several classes defined in this Project (and, thus, in every Dataset): a building, a traffic light, a vehicle and so on. All classes are set to the "bitmap" shape, so that there is no way someone will accidentally create some "Cars" with polygon tool (a set of points), and some "Cars" with bitmap tool (a set of pixels).

Also, there are two tags defined: a "vehicle_type" of several pre-defined options ("Bus", "Bicycle", "Train") and a "color" that accepts any string value.

Datasets are used to split data into a "subfolders" to make data management easier. For example, you can then define a Labeling Job to label all Vehicles in Zurich.