Labeling
  • 05 Aug 2024
  • 2 Minutes to read
  • Dark
    Light
  • PDF

Labeling

  • Dark
    Light
  • PDF

Article summary

This article applies to these versions of LandingLens:

LandingLensLandingLens on Snowflake

After you've uploaded images to LandingLens, you can label them. Labeling images is an important step of the model creation process. The labels you provide on images are called ground truth. When your model trains, it learns to detect objects of interest by referring to the ground truth, so it's important that your labels are accurate and consistent.

Tip:
Create a label book to establish clear labeling guidelines and provide proper training to help labelers understand your project's requirements.

Label Images

Note:
Before you can train a model, you must have at least 10 labeled images for Object Detection and Classification and two labeled images for Segmentation.

To label images:

  1. Click to open the image you want to label.
  2. Perform one of the following actions:
    1. If the image has no objects of interest to detect, click the No Class button. LandingLens marks the image as No Class.Mark the Image as "No Class" When There Are No Objects of Interest
    2. Use the provided labeling tools to label the object of interest. LandingLens automatically saves the label.
      Example of Labeling with a Bounding Box

For more information on labeling images using specific labeling tools, refer to the relevant article based on your project type:

Remove Labels From Images

In LandingLens, there are multiple ways to remove labels from images. For example, you can quickly remove all labels from an image and then relabel it. Or, you can mark an image as "No Class" to remove all labels from that image.

For more information on removing labels from a specific project type, refer to one of these relevant articles:

  • Object Detection 
  • Segmentation

    Caution:
    You cannot undo removing labels. However, you can create a snapshot of your data as a backup if you want to remove all labels. For more information on dataset snapshots, go here.

Remove All Labels From Images

Let's say you recently added new classes to your project and need to rework the labels on your images. You can quickly remove all labels from an image. To do this:

  1. Click to open the image you want to remove labels from.
  2. Click the Remove All Labels icon.
    Click the "Remove All Labels" Icon
  3. A pop-up window displays. Click Remove All Labels in this window to confirm the deletion of the labels. LandingLens automatically removes all labels from the image.
    Confirm to Remove All Labels 

Remove All Labels and Mark Images as "No Class"

Let's say you accidentally labeled an image, but there are no objects to detect in that image. You can quickly remove those labels and mark the image as No Class. To do this:

  1. Click to open the image you want to edit.
  2. Click the No Class button.
    Click "No Class" to Remove Labels and Mark the Image as "No Class" 
  3. A pop-up window displays. Click Remove All Labels and Mark as No Class to confirm the action. LandingLens automatically removes all labels in that image and marks it as No Class.
    Confirm to Remove All Labels and Mark the Image as No Class



Was this article helpful?