Classification
  • Dark
    Light
  • PDF

Classification

  • Dark
    Light
  • PDF

* Available on image and video items

The Classification tool enables labeling/annotating an entire image or scene (in the case of a video file). For example, classification can be used to classify product images under categories, subcategories, and characteristics, such as men’s clothes, polo shirts, etc.

The Dataloop system has several important features designed to increase the productivity of classification annotation tasks.

Classification in the Dataloop JSON Format
{
	"type": "class",
	"label": "labelName",
	"attributes": [],
}

Image Classification

recipe

  1. Go into the annotation studio.
  2. Select the Classification tool.
  3. Select and click a class/label from the labels list (label picker) to create classification annotations.
  4. The newly created annotation can be seen
    A. In the annotation list.
    B. As a label chip at the top of the image.

Auto Next Item

recipe

When the task requires only a single class per image, enabling the “Auto next item” feature will automatically load the next item as soon as a class is selected, to save time.

Auto Complete

Enabling “Auto complete” (shown only after ‘auto next item’ is checked) will automatically set an image item to ‘Completed’ status when moving to the next item.

Bulk Classification

Rapid and highly efficient item (image/video) classification can be done from the dataset browser in bulk:

  1. Go to the tasks page and double click a task.
  2. Click on the “Browse task” button.
  3. Adjust your view with thumbnail size and number of items per page to fit your needs.
  4. Select items (CTRL+Mouse click) you want to classify.
  5. Click on the classification tool icon
    image.png
  6. Select a class from labels list, and approve the action by one of hte 2 options
    1. Classify - Simply adds the class to the item, but does not change its status
    2. Classify & Set Status - Adds the classification and opens a drop-down to select which status to add to the item (e.g COMPLETED by default, unles custom statuses were added to the task).

image.png

Baluk Classification Limiations

Bulk classifciation requires that all items are allocated to an assignment, and therefore it disabled when the task is a "Pulling" based task, where items are pulled by assignees during work, and not in advance. Use only "Allocation" tasks for bulk classification work.

Video Scene Classification

Classifying a video scene is similar to how images are classified, with the main difference that a scene has Start and End times.

  • To classify scenes in a video, navigate to the first frame and click on the label. The annotation will be created with the 'START' keyframe and last until the very last frame of the video.
  • Continue skipping frames until you reach the end of the scene and set the frame as the end time to conclude the annotation. See how to set the end time here.
  • Alternatively, if you want to use that label again, you can simply hide it for a number of frames and unhide it once you need it again. Learn more about hidden frames here.

hidden annotations.gif


What's Next