February 2022 Update

February 2022 Update

February 28, 2022
A series of cards from V7 Darwin

Introducing: The New Model Tool

Auto-Annotate is getting a major power-up: Use any model to automatically label polygons, bounding boxes, text, and tags. The Model Tool lets you map classes to models from the internet, and will surface confidence scores for every label created.

Confidence scores within the Annotation interface

See confidence scores for every annotation, spot ambiguous objects and set confidence scoring thresholds to control the accuracy of your automatically-labelled data. 

Support for all model types 

Whether it's one of your team's trained object detection or classification models, or one of our public models like the text scanner, the Model Tool can output any annotation type. You may even use it at lower complexity levels, for example use an instance segmentation model to produce bounding boxes.

Label specific regions or entire files

Create multiple unique labels at once by drawing a bounding box over several objects, or run a model over an entire file by hitting Cmd + Enter.

Model Tool also works on ultra-large files, medical imaging, and PDFs.

Triage hard tasks with AI Models 

One of our favorite ways of using Model Tool is to pre-process data and flag low-confidence boxes. This is a great way to flag out-of-distribution objects and direct reviewers to them. You may also use multiple model versions to compare their score on the same image.

New Features

🆕 Label segments can now be easily moved/resized in video timelines when annotating on touch displays.

Performance Improvements

🚀 We've made zooming and panning in multiple-gigabyte TIFF files significantly smoother.

🚀 DICOM series and their slices now have significantly-improved loading times and playback speeds.


👾 New annotations now always appear at the top of the Annotations menu.

👾 Corrected the alignment of rendered frames in streamed videos with the frame in the video timeline.

👾 Fixed an issue that prevented Auto-Annotate from functioning in videos with streaming enabled.

👾 Corrected the interpolation algorithm for Ellipse annotations to ensure smooth interpolation between key frames.

👾 Fixed an issue that caused the layering of annotations in the Annotations menu to be inverted when compared to the layers in the exported JSON file.

👾 Tweaked interpolation behaviour to prevent complex polygons (polygons with multiple parts) from disappearing in interpolated frames.

👾 Files in a Complete state no longer display any annotation tools to reflect the fact that they are uneditable.

👾 Fixed an error that prevented DICOMs with hanging protocol layouts from being edited.

👾 Class filters now reset when moving between datasets.

👾 All class names were previously capitalised in the workview. This is no longer the case.

👾 Keypoints now change appearance when hovering over them with the cursor. This makes them clearer when not the cursor is not hovering above them.