Labeling Tools

Learn about the labeling tools that are available to you in Scribe.
Clarifai Scribe: Labeling for Bounding Box Detection

Scribe provides special tools for working with images and video.

Easily label images with bounding boxes


Brightness, saturation and color inversion

invert colors

You can enhance the visibility of your photos with image adjustments. Image adjustments can be combined. Just click reset to return to the original version of your input.

Zoom and pan

pan image

Powerful zoom and panning features allow you to closely inspect specific regions of an image. Just click reset to return to 100% zoom.


General Keyboard shortcuts

  • Left arrow - Previous input

  • Right arrow - Next input

  • Enter - Submit label

  • "/" - Filter Concepts

Image Keyboard shortcuts

  • B - Brightness

  • S - Saturation

  • I - Inversion

Polygon Labeling

Many labeling tasks can be handled well with bounding box labels, but in cases where you need a more precise way to annotate object, polygon labels are an excellent option. Polygon labels allow you to identify and annotate the exact pixels of your image that represent the object that you would like to label. Polygon labels will output a sequence of x, y coordinates for every point that comprises the polygon.

First you will need to select "Polygons" as the task type when creating your labeling task.

Select "Polygons" as your task type when creating a new labeling task

When labeling your images, you will be able to create multi-point shapes that can outline the precise pixels of the object that you would like to label. Just remember that you will need to "connect the dots" by connecting the last point in your polygon with the first point.

Outline your image and be sure to connect the first and last nodes of your polygon


Scribe provides powerful tools for labeling video. When working with video, you can leverage video interpolation tools to quickly label thousands of individual frames of video. This rapid labeling technique makes video an excellent source of training data, even if you want your model to primarily analyze still images. You can QuickTrain and DeepTrain models on video data that has been labeled in Scribe.


Interpolation allows you to quickly label multiple frames of video with the same concept. Simply select the interpolation icon and draw a bounding box around the object that you would like to label. Then scrub the video player to a new point in the video and move and adjust the bounding box to the new location of the object. Interpolation will automatically draw a series of bounding boxes between them and a keyframe marker will identify the point in the video where a new interpolation begins.

Track Suggestions for AI-Assist

AI-Assist can automatically track objects across multiple frames of video. Just be sure to include a "centroid-tracker" in the workflow that you are using for AI-assist. Separate instances of a given concept will be detected and versioned for labeling. You can even use the timeline editor to scrub back and forth between different frames of video.

Video Keyboard shortcuts

  • Q - Start of video

  • W - Scrub backward

  • E - Scrub forward

  • R - End of video


Scribe makes it easy to label your text data. You can review text inputs in the same view that you would review images for classification tasks.

Label text data in Scribe

Editing Labels

You can easily resize, move, or relabel existing bounding boxes. Just click on the bounding box that you would like to change and drag to move or resize. Bounding boxes can be relabeled by clicking the "edit" icon next to the existing label in the right hand sidebar.

Quick-Mask Tool

The quick mask tool makes it easy to select image masks when creating polygon annotations. Just be sure to select the "polygon" Task Type when creating your labeling task. The Quick-mask tool allows you to select a region of an image, and then will automatically generate an images mask for the detected object in the image. You can also fine-tune the selection once a mask has been created.