The Dataset page now has a filtering and pagination system for a better user experienc
Tasks are now sorted by image name for default (for example in the task-list table under project or by start labeling)
If you connected your Dataset with an Amazon S3 Bucket you are now able to sync more than 1000 objects
We implemented our new nested geometries feature. It allows you to create a child-geometry to have more context.
You can navigate through your created objects with the STRG+ARROW key. It will help you within the review process.
Image segmentations of the same type are now grouped
The toolbar user experience was improved. You now can reset the sliders for contrast, staturation and brightness individually
We also improved the workspace info dialog by grouping informations into different tabs
Navbar: We improved the user experience by rearranging the elements in our left navigation bar
We have improved the performance of the import a lot
Negative coordinates are now forbidden
The dashboard accidentally counted even deleted images
The reviewed_skip task-state is now also a part of the dashboard charts
We implemented our new image segmentation feature. It allows you to partition the whole image in multiple segments
The image segmentation comes with a bitmap export for each individual class
The image segmentation data will also be a part of the export functionality
Improvements for the import feature:
It delivers more useful and informative error messages
We simplified the json import by reducing the required fields
The project task-list now shows badges which indicates wether the task was pre-labeled and imported
Administrators now have the power to unassign labelers from a task
Workspace: Moving polygon points wont stick to a virtual grid so you have more control about it
Label configuration: You now can use question marks for the field description which can help the labeler understand the task better
Workspace: Objects of the type line now cant be drawn outside the workspace area
Workspace: Duplicating geometries now works even the geometry is very close to an edge
In your task-overview you now see the individual usernames instead of the user-ids
The review-mode now also shows skipped tasks with an additional indicator for the reviewer
The workspace task-control buttons now show an clear text instead of icons
We overall make it easier to click action-icons like the delete button
With the shortcut CTRL+V you can now copy a geometry and its classifications to speed up your daily work with DataGym.ai
The AI-assisted pre-labeling now supports also polygon geometries
The line geometry appears now like a polyline. Untill now a line was defined with the starting point and end point. But now you can use multiple points and also add points like on the polygon.
In addition the helping points are removed from the polygon geometries. Instead, an addition point can be created just by hovering and clicking over the line between two corner points of the polygon.
To support you on the review process every geometry can now be commented.
Simplified the configuration for the AI-assisted pre-labeling.
Changed the calculation for the AI-assitsted labeling if a polygon should be replaced with a new one.
On fastly using the shorcuts the task may could not be completed.
With our review feature you can comment and reject the full task. With this release you can also comment single geometries within the Workspace of DataGym.ai.
Updated the front end of our pre-labeling feature introduced with the last release.
Removed the hint to the 'comming soon' benchmark tool.
Some rewordings within the application and the webpage.
Fixed the points selection tool for our AI assisted labeling. Fast right clicks produced sometimes positive points instead of negative points.
Fixed and simplified the API importing tool.
Let DataGym.ai do the boring tasks for you and get the bounding boxes from the brand new pre-labeling feature.
In order to priorize other features we deactivated the benchmark tool. It will be available later.
Now it is possible to hide single geometries not only groups.
Fixed some bugs within the workspace.
Our latest release comes with a set of brand new tools for AI-assisted labeling. Speed up your labeling process by using a variety of tools, such as Paint fill, Rectangle selection, and Click. You can also now refine the predicted labels by setting additional points in your image. Check out our documentation for a complete overview.
Measure the labeling performance of your team members with the new Benchmark Mode. This feature allows you to review the accuracy of your team members while labeling images. The Benchmark Mode is still under development but will be fully available soon.
Delete the images in your datasets faster with multi-selection.
Use our new dynamic cancel button in the label mode to cancel or deselect your current action. Thereby, you can now work on your labeling tasks without using keyboard shortcuts.
Added a warning message to your image list when an external image is not reachable or uses an invalid data format.
Hide helping points while editing polygon geometries when points are too close to each other.
Copy existing geometries or image classifications within the label configuration. The copy got a new tag, export key and in case of a geometry also a new color and if possible also a new shortcut. All the nested classifications are reused to speed up the configuration job.
Removed classification types radio & select. Both are replaced with the new type option to simplify the label configuration.
Removed also an error message when the last task was processed. Now you get a success message because all your jobs are done for the moment. Enjoy the day.
The preselected color for new geometries within the label configuration rotates.
Some UI/UX improvements.
Some translations were changed.
Set boundaries to the drawing mode so geometries cannot be moved outside of the image.
Add a context menu to label mode so you can change the geometry type if necessary. Read more within the documentation.
Use the review feature to raise the quality of your labels.
Added some statistics to the project overview.
Simplify the import Python api.
Finish polygons also via double click or by selecting the first point.
Some minor changes to make your life easier.
Fixed the closing action on upload images modal.
Some bugs within the JSON API.
Add automatically a dummy-project to your personal account so you can quickly learn how DataGym works.
Use the new JSON-Import to upload label predictions to DataGym. Instead of labeling images from scratch, correct just the pre-labeled images. Don't waste your time with boring tasks.
Get an overview about your limits within your account page to not raise any restrictions.
Some minor changes to make your life easier.
Inverted the arrow key direction to move the image within the label mode.
Fixed an error within the AI-assisted labeling
Now you can upload images to datasets
Within the projects you can setup your label configuration
Connect projects with datasets
Manage your tasks