Changelog

April Update

April 24, 2020
A series of cards from V7 Darwin

New Image and Video Cards

You can now visualize image filenames, numerical ID, dimensions, filesize, annotation status, and multiple tags right from the image card. An improved card-size dial now expands the cards up to a full-screen size, and allows you to view more tags at once. Hovering over an image reveals additional info, as well as more tags. Delivery of these cards has also been sped up significantly.

Easy Instance IDs - Track objects through images and video

Within Darwin, each new annotation is treaded as an independent instance. Until now though, you couldn't track these objects in between frames without using a text ID or attributes. Instance IDs allow you to copy over the object IDs of annotation between frames, or within the same image. These can be carried over in bulk between images with the Copy Instance button, and can be pasted to any annotation with CTRL+Shift+V. They are also chroma hashed, to easily distinguish them visibly by a unique color trio when numbers become large and challenging to remember. They will work up to 1 billion unique instances per dataset. Try them now by adding them as a sub annotation to classes.

Auto-Annotate Controls!

You can now edit, re-run, re-shape, and copy+paste Auto-Annotate objects. Your editing clicks and initial box are saved automatically so you can change these during review or at any time in the future. You may also copy over objects between images and re-run Auto-Annotate to adapt them to their new appearance. Graphical notes to signify a loading click have also been added.

Drag a line, make a point

You no longer have to click to generate vertices on a polygon, and later move them. You can now simply drag way a new point!

Export Selected and Annotator Metadata

Pick any set of filters you wish, press "Select All", and generate a custom dataset version export that matches this combination of filters. This way, you can also export incomplete or archived images. Annotator metadata is also available upon export, adding authorship information to each annotation.

A New Content Delivery Network (CDN)

This one is for everyone working from home! That probably includes you, the reader. Over the past weeks we set up a new CDN system to improve Darwin's performance in several regions where our users are located. With global bandwidth being massively increased, it's evermore important for us to focus on an optimized platform that works all over the world. Non-sensitive files, such as interface elements, are now stored closer to you geographically, while sensitive information such as image data remains in our secure Ireland-based servers, yet are pre-loaded more cleverly to reduce loading speeds.

More Features and Improvements

Feature
You can now sort datasets by any combination of multiple filters belonging to different data columns.
Feature
The naming of annotaor report exports has been changed from report.csv to the appropriate team and dataset name, and sampling rate.
Feature
Feature
Performance
Performance
API keys are now shown in mono font for readability
Performance
Original filenames are now visible on hover on images in carousel
Performance
Ultra-high res images (above 10,000 pixels squared) now respond more quickly to panning.
Fix
Fixed a bug that would direct user/admins to “Annotator Mode” when reopening Darwin after a 10 minute period of inactivity
Fix
‍Fixed a bug that would show a white screen and not load any content when reopening Darwin after a 10 minute period of inactivity
Fix
Users can no longer spam-click a tag causing it to be applied multiple times during connection delays.
Fix
Unassigned/author-less tasks now load images correctly and don’t get stuck.
Fix
Comments now display the box where the comment is applied on hover, not just on click.
Fix
Fix

Minor Update: Early April features and fixes

April 6, 2020
A series of cards from V7 Darwin

More Features and Improvements

Feature
Export Selected: You can now apply any filter to a dataset and export only the selected images.
Feature
Annotator Export Metadata: You may choose whether to add annotator authorship to each annotation in an export. This option is available in the export dialog.
Feature
Workview back button now works if an image is accessed via a link.
Feature
Performance
Token stability improved: Users should be able to keep Darwin open on multiple tabs, and receive less errors related to timers and any real-time operation.
Performance
Toast messages (those appearing at the bottom of the interface) are more persistent and re-appear if an error is not resolved.
Performance
The layout of the notification page and icon has been improved to fit more browser resolutions.
Performance
Fix
Dataset management now loads more than 500 images when multiple filters are applied.
Fix
Rendering annotation order (how they overlap from front to back) would sometimes display incorrectly and has been fixed.
Fix
More than 9 notifications can be seen after switching to another team.
Fix
Bugs on the tutorial page related to sub-annotations, auto-annotate, and layer ordering have been fixed.
Fix
The class and attribute color selector should no longer squish into a square.
Fix
Fixed a bug that caused image processing to halt during large video uploads.
Fix
Fixed text mis-alignment in certain error pages.

March Update

March 19, 2020
A series of cards from V7 Darwin

New Stats!

Head over to the Annotators section. Previously you could only see current image statuses. Now you will be abe to see the pass/reject rate over time as well as the overall dataset progress.

Sorting within datasets

There's a whole new sidebar in Dataset Management. At the top, you will see sorting options for the current dataset.

New status filters

We trimmed the fat on status filters. They are now fewer and more intuitive to use.

Tag with one click, and filter more easily.

Tagging a bundle of images previously required 3 or 4 clicks, and we cut it down to 1. You can also visualize the number of annotations at a glance, filter through them, and create new tags on the go without having to go through the class-creation system.

Reset Zoom, or Not.

Most of the time, switching to the next image means wanting to center the image on your window (Tip: the hotkey for this is 0). However sometimes you will want the zoom to be preserved, such as when you are annotating one part of a video, to track an object. You can now lock zoom when switching images in workview with a button on the bottom left.

More Features and Improvements

Feature
XML Exports are now available.
Feature
Cuboid annotations are available by default for all teams.
Feature
Feature
Performance
Error toast messages are more persistent to prevent from annotating without a connection.
Performance
Performance
Performance
Fix
The Upload button no longer moves around when completing an upload.
Fix
Fix
Fix
Fix
Fix
Fix

February Update

February 28, 2020
A series of cards from V7 Darwin

Image Manipulation

One highly requested feature is to allow manipulating the image's appearance. This means tweaking contrast, saturation, and the visibility of annotations.

You can now tweak these through an Image Manipulation panel on the top right. We will keep adding more ways to tweak images, including color channels and image overlays. If there is something you'd like to see when it comes to image manipulation let us know!

Video Folders

Video frames are now bundled as folders and are easier to browse. This is a first step towards scrubbable videos which would enable faster frame-by-frame annotation.

Smoother Large-File Uploads

We are seeing larger and larger uploads coming through. We've smoothed out the progress of uploads so it's easier to keep track of.

More Skip Controls and Mark as Done

As requested by some of you, we added more ways to skip images, as well as ways to manually mark un-changed images as done.

Polygons close magnetically

A small tweak but a highly requested one. Next up are better ways of manually drawing masks.

More Features and Improvements

Feature
Assigning an annotated image now carries over annotations into the new task.
Feature
Annotation metrics of users that haven't labeled a dataset are hidden.
Feature
Images are automatically rotated when Exif data requires it, such as direct uploads to Darwin from mobile apps.
Feature
Performance
Improved the time tracking reliability.
Performance
Database tuning is now performed daily.
Performance
Reduced the number of DOM elements in workview to scrolling performance.
Performance
Fix
New dataset export notifications are now clickable.
Fix
Open Datasets now display attributes.
Fix
The Auto-Annotate functionality in tutorial no longer disappears for certain users.
Fix
Deleted attributes no longer continue to stay in annotations.
Fix
The Z-Indexing (annotation ordering) in workview now correctly reflects the inverse .Json ordering for all annotation types.
Fix
Fixed a bug where assigning images with too many existing annotations caused crashes.
Fix
Names of deleted datasets can now be re-used for new datasets.

January Update

January 20, 2020
A series of cards from V7 Darwin

Track the progress of your labeling workforce

Late last month the Annotator tab became available, allowing you to monitor the performance of labelers across all datasets with granular detail.
We based this feature on your feedback, adding the ability to monitor:
- Images completed
- Labels completed
- Total time spent labeling
- Labels per second
- Accuracy score (based on review)
These metrics can be filtered by day, week, or month and are beautifully plotted on a comparable graph.

Version Datasets as you Export

Version control in machine learning is a bit of a mess. When exporting a dataset or any batch of labels, Darwin will create and save a version of those labels for future re-use should the dataset change over time.

APIs for Everything

For teams unrestricted by GUIs. There's a lot you can do with APIs now: fork datasets, create classes, download reports, and a lot more. Try the new API key management or visit the new version of Darwin.py at
https://pypi.org/project/darwin-py/

Navigate

We're looking out for ways to reduce fatigue and save time. You can now pan images with the WASD keys, use the middle-mouse button to pan during Auto-Annotate editing, and you'll find scrolling to zoom has become smoother.
Several more quality-of-life improvement like these are around the corner.

Neural Networks Beta

Those of you who met with our team in recent days may have had a sneak-peek at the automated model training and deployment system being developed. It's intended to speed up the federated learning of Auto-Annotate, and enable further labeling automation.

More Features and Improvements

Feature
Clicking on an annotation in workview scrolls to its position on the sidebar and highlights it.
Feature
You may now hide sub-annotations as well as all labels. Hiding has been sped up significantly.
Feature
If you minimize the sidebar it will be remembered as a preference.
Feature
.TIF file support was expanded further.
Performance
The image carousel in workview has an improved pagination feature to support 1M+ images.
Performance
Improved error toast messages throughout the platform.
Performance
Image download speed via CLI has been greatly improved.
Performance
Switching teams has been sped up.
Fix
Archived images can be permanently deleted by users and admins.
Fix
Team avatars no longer appear squished if the aspect ratio is very uneven.
Fix
Fix
Fix
Fix
Fix
Newer
Older