DocumentationRecipesReferenceGraphQLChangelog
Log In
Back to All

March 9th, 2022

about 2 years ago by Kili Technology

Changelog (v2.107.0)

New

  • [Labeling] To speed-up data extraction tasks in PDF projects, we now provide the ability to onboard optical character recognition (OCR) metadata before labeling. For details, refer to Optical character recognition.
  • [Labeling] Within a frame project, you can now extend the navigation slider to navigate and label up to 500 frames at once.
  • [Assets upload] You can now delete the notifications displayed when errors are encountered during asset upload.
  • [Queue] If you’re using the free “discovery” plan to access Kili, you can now use the Explore feature.
  • [Queue] When reviewing asset labels, you can now view all labels of an asset directly from the project queue. Previously, it was only possible via Explore.

Change

  • [Labeling] We reinforced our rules regarding autosave during the labeling phase. All of your work is now saved, even if you leave the labeling interface before the automatic autosave. Thanks to those changes, you will never lose ongoing work anymore.
  • [Labeling] When you create a semantic segmentation project or add a new semantic segmentation job, the semantic smart tools (interactive segmentation and superpixel) are now visible by default on the toolbar of the labeling interface.
  • [My Organization] In “My organization” page, you can now view the information on metrics and organization members in separate views. This makes the interface more readable.
  • [Analytics] In the “Analytics” page, you can now view the information on progress, project annotations and project labelers in separate views. This makes the interface more readable.
  • [Analytics] In the “Analytics” page projects that were made public, labels made by users from outside of the project are now marked as “Public user” instead of “Anonymous”.
  • [Queue and Explore] Up to now, project members could see all auto-saved labels (including those made by other project members) which may have caused confusion. Now, project members only see their own auto-saved labels.
  • [Assets upload] Error reporting when uploading assets has been redesigned and is much clearer now.
  • [Labels export] When exporting labels to YOLOv4 format, classes are now listed in a separate classes.txt file. Instead of specific class names, annotation files contain indexes to these classes.

Fix

  • [Labeling] In pose estimation projects run on the Microsoft Windows operating system, we fixed a bug that didn’t allow for using the pinpoint tool with extra precision.
  • [Labeling] In frame projects and image projects, we fixed a bug that prevented displaying some annotations in Explore view.
  • [Labeling] In image transcription projects using OCR, we fixed a bug that caused transcriptions from other documents to show up on screen, instead of proper ones.
  • [Labeling] In image and frame projects that contain ML-based predictions, we fixed a bug that caused new annotations to be displayed as ML predictions.
  • [Labeling] In relation tasks, we fixed a bug that didn’t allow for hiding relations from the asset viewer.
  • [Assets upload] We fixed a bug that did not allow for importing remotely-hosted assets with duplicated IDs. When Kili finds a list of duplicated assets in the CSV file, only the first of the duplicated assets is inserted; the other duplicated assets are rejected without causing an import error.
  • [Project settings] We fixed a bug that prevented anonymous users from using the public labeling interface.