Now you can label your images for object detection directly within Roboflow.


Roboflow Annotate is a self-serve annotation tool included with all Roboflow accounts that greatly streamlines the process of going from raw images to a trained and deployed computer vision model.
Whether you need to correct a single annotation or label an entire dataset, you can now do it within Roboflow without having to download a separate program or go through the process of exporting and re-importing your images.
Annotating images in Roboflow is simple.

How to Access

The self-service version of Roboflow Annotate has been added to all existing plans for no extra charge. Simply click on an image in one of your datasets to open the labeling interface. All changes will automatically be saved and shared with the rest of your team.

The Labeling Interface

On the right-hand side of the labeling interface, you will find the toolbar. It contains the following options:
  • Select (represented by a hand icon) allows you to select, edit, and drag individual annotations.
    • Single-click an existing bounding box to select it. Once selected, you can change a bounding box's size with the circular white handles that appear on its corners and on each side. Or use the class editor to change the box's label.
    • Drag a box to move it.
    • Drag the background to pan.
    • Click the background to deselect all boxes.
  • Create (represented by a rectangular box icon) allows you to draw new annotations. In this mode, you will see crosshairs that will help you determine where to start drawing.
    • Click and drag across the image to create a new annotation then use the class selector to choose its label.
  • Undo reverts the previous action.
  • Redo reverses a previously undone action.
When an image is selected, the Class Selector will appear. It contains the following options for choosing the label of a bounding box:
  • Textfield to create a new class or filter existing classes.
  • Buttons to save or discard your changes.
  • Class List of the existing classes in the dataset (filtered by the text field and with the active option highlighted in purple) and, sometimes, a "Create class" option if the text you typed does not match an existing class.
At the bottom left of the screen, you will find the zoom tool. You can zoom in and out to fit more of the image on your screen at one time or to get a closer look for more detailed editing.
On the left-hand sidebar you will find the following drawers:
  • Annotations (abbreviated Annots) show which classes are present and not present in an image and what color their boxes are. (Watch this spot for updates soon as we add filtering and fine-tuning features.)
  • Attributes contains information about your image including its dimensions, last-modified time, and whether it is in this dataset's training, validation, or test set.
  • Raw Data is mostly used for debugging and can be safely ignored by most users unless our support team requests that you access it. It contains Roboflow internal representation of the image.
In the header you will find navigation to exit the labeling interface, move on to the next or previous image in the dataset, and perform the following actions:
  • View source image (represented by an eye with a slash through it) temporarily hides the bounding boxes to give you a closer look at the original image.
  • Options (represented by an ellipsis)
    • Use as Cover Photo marks the current image as the icon for the dataset on your homepage and uses it as the preview image for preprocessing and augmentation options.

Keyboard Shortcuts

The following keyboard shortcuts are available to speed up your labeling flow. The meta key is usually the command key on macOS and the ctrl key on Linux and Windows.
If no annotation is selected, the following keyboard shortcuts apply:
  • meta - temporarily switches between the select and create tools while held down. (For example, if you are in create mode, hold down the meta key and click an existing box to select it.)
  • c - switches to the create tool.
  • d - switches to the select tool.
  • plus - zooms in.
  • minus - zooms out.
  • zero - returns to default zoom (fits image into the viewport).
  • one - zooms to 100%.
  • escape - exits the labeling tool.
  • left arrow - navigates to the previous image.
  • right arrow - navigates to the next image.
Once an annotation is selected and the class selector is visible the shortcuts are as follows:
  • enter - saves the active option (highlighted in purple) as the current label for the selected bounding box.
  • escape - cancels and deselects the current box without changing its label (if the currently selected box was just drawn it will be deleted).
  • up arrow - changes the active option to the previous one.
  • down arrow - changes the active option to the next one.
  • backspace - deletes the current bounding box if there is no text in the text field (this means you will usually have to push backspace twice to delete a box after selecting it; one time to delete the highlighted text and a second time to confirm deletion).

Best Practices

We have gathered a collection of image labeling best practices that will avoid common pitfalls down the road. We recommend reading it before starting to label your dataset.

Model-Assisted Labeling

In February 2021, we released model-assisted labeling. Model-assisted labeling allows you, with one click, to have a model take a first pass at annotating your images. While we strongly encourage that a human reviews the annotations to make sure the labels adhere to best practices, this feature significantly speeds up your team's annotation efforts!
Our Sandbox, Growth, and Enterprise accounts support using the predictions from your Roboflow Train models, or our trained version of the Microsoft COCO 2017 Dataset for Object Detection as a starting point for annotating additional images. Roboflow Public Workspaces can leverage Label Assist after the addition of training credits to the workspace.
Please contact our sales team to upgrade.
Last modified 3mo ago