Workspaces, Projects, and Versions

An introduction to workspaces, projects and dataset versions on Roboflow.

Written by Mohamed Traore

Last published at: April 20th, 2022


All Roboflow projects belong to a workspace.  When you sign up for a Roboflow account, we automatically create an initial workspace for you.

Workspaces are how you can manage who has access to what projects and what features are available for projects inside the workspace.

Workspaces contain Projects which contain Versions that have trained models and can be exported in various Formats.

Creating a New Workspace

To create a new Workspace, simply click the "+ Add Workspace" button on the nav menu after you log into Roboflow.

When you create a new Workspace, you need to choose a plan for the Workspace.  To learn more about the available plans you can check out the Plans and Pricing section.

Adding Team Members to a Workspace

To invite a team member to a Workspace, add their email address in the Workspace Members section of the workspaces settings.

Workspaces can be created or deleted. Workspaces are irrecoverable once fully confirmed as deleted.


A project lives within a workspace and houses a dataset of images that are: fully labeled, partially labeled, or unlabeled. 

Projects can be merged, duplicated, created, or deleted. Projects are irrecoverable once fully confirmed as deleted.

Creating a New Project

Select "Create New Project." This will trigger a model to pop up with an option to upload your own data or go through a tutorial with sample data.

Upload your own data or download sample project tutorial.

Upload your own data or download sample project tutorial 

Roboflow currently offers dataset upload, labeling, dataset generation, and custom model training support for the following project types: Object detection (bounding box), single-label classification, multi-label classification, and instance segmentation.

Roboflow currently offers dataset upload, labeling, dataset generation, custom and AutoML model training, and deployment for the following project types: Object detection (bounding box), single-label classification, and multi-label classification.

Selecting "Upload Your Own Data" requires three fields to be passed in:

Project Name

A way to refer to your collection of images/videos.

  • If you're uploading a bunch of images of chess pieces, you might name this "Chess Data."
  • The dataset name must be unique among your datasets. (For example, you cannot have two datasets both named "Chess Data.")
  • Right now, we do not support editing the dataset name once you have created the dataset. If you must edit your dataset name, you can re-upload your data with the new name or contact us.

Project Type

  • Single-Label Classification: A good rule of thumb for when to use object detection vs classification is whether the things you're trying to predict are "objects in an image" vs "properties of an image".
    • For example, a chess piece is an "object in an image", but winter is a "property of an image". If you were trying to draw a box around the winter or daytime part of an image, you'd likely end up drawing a box around the whole thing.
  • Multi-Label Classification: Similar to Single-Label Classification in terms of finding "properties of an image", only multiple properties of an image.
    • For example, If you were trying to detect not only winter, but the day, cloudy, and night as well on the same image.
  • Object Detection: Useful if you are attempting to identify one or more objects in an image with bounding boxes. A good rule of thumb is if the object will need to be detected in motion or in position.
    • For example, a chess piece moving from one square to another, recognizing whether or not the chess pieces are where they belong on the board during the time of set up. 

If you can't decide, we recommend starting out by labeling your images for object detection, because while you can convert an object detection project to a multi-label classification project easily, to convert in the other direction will require re-labeling your dataset.

  • Instance Segmentation (also known as image segmentation): Useful for when you need to measure the size of detected objects, cut them out of their background, or more accurately detect oblong rotated. With instance segmentation, your application can determine the number of objects in an image, the classifications, and their outline.
    • For example, if you need to measure the size of a tomato leaf in order to remove it from its background, or to measure a lawn from satellite imagery.

Note that instance segmentation models are typically larger, slower, and less optimized for edge deployment. Instance segmentation models may need bigger datasets to obtain the same accuracy as object detection models.

Other Project Types
Roboflow currently does not offer support for other project types, however, requesting early access is highly encouraged. Below are some of the different project types that will be supported in the future and when to use them. You should only use instance segmentation if the specificity of the object's outline is required by your application.

  • Semantic Segmentation: For attempting to identify multiple objects of the same or different classes in images with freeform polygon shapes (not bounding boxes).
  • Keypoint Detection: For attempting to identify the locations of important components in an image

Annotation Group
This should be the broader class of objects being detected or the collection of categories for a classification problem. It is a way to refer to all of the objects or labels in images.

  • For example, when attempting to identify pawns, rooks, kings, and queen pieces on a chess board, the annotation group can be pieces.
  • Or, when attempting to classify handwritten images as being 0, 1, 2, ... 9, the annotation group can be digits.

To learn more about annotation groups, read our blog titled What the heck is an annotation group?


A dataset or project version is a point-in-time snapshot of your dataset.

By keeping track of exactly which images, preprocessing, and augmentation steps were used in each iteration of your model you maintain the ability to reproduce the results and scientifically test across various models and frameworks while remaining confident that the results are attributable to the model changes and not due to a bug in the data pipeline.

Versions can be created or deleted. Versions are irrecoverable once fully confirmed as deleted.