Alongside that, this release adds a confidence threshold slider to Find Similar, contextual back links on dataset pages, and fixes several workflow issues including region counts after undo/redo, form validation visibility, and a critical Delta Layers MultiPolygon problem.

Changelog​

New Features and Improvements

Fixes

Deep Dives​

Action History and Tasks​

Our Action History system has been improved to better show progress of compatible tasks.

When you perform an action that uses tasking (like pasting objects), the Action History entry will now show a progress indicator, as well as some new controls that would otherwise not be there.

This does not quite look like an ordinary action entry. You can see the icon is replaced by a progress spinner, and the "Retry Action" button is linked to the task system instead of simply redoing the entire action. This allows you to retry only failed subtasks, which should avoid duplicating work that has already succeeded, and overall result in a much smoother experience when performing large actions that involve many objects.

Beyond that, if you'd like to see more details about the task, you can click on the Details button to see the Task Details view, which will show you all subtasks, their status, errors, and even allow you to retry individual subtasks if you'd like.

Changelog​

New Features and Improvements

Changes

Fixes

GeoSense is our in-house assistant for geospatial data handling in the form of a chatbot with specialized tools to query your data and get insights.

This update also comes with some important fixes and improvements to the region creation experience. We have added new region validation criteria to make sure you organize your data in ways that lead to optimal model training results.

Changelog​

Major New Features

New Features and Improvements

Fixes

Deep Dives​

Region Creation Feedback​

When creating inference or training regions, users will now get dynamic feedback on the validity of the region they are drawing. This includes:

• Regions with any dimension spanning less than 512 pixels will be flagged as "Too Thin".
• Training Regions with an area larger than 50 Megapixels will be flagged as "Too Large".
• Training Regions that occupy less than 50% of their bounding extent, or inference regions that occupy less than 25% of it; will be flagged as "Too Sparse".

This feedback runs in two stages: on creation (both dynamically and on submit), which enforces the rules above, and on usage. This means that users will also see warnings when trying to infer on regions that don't meet the criteria or have generated warnings, or when trying to train on any dataset involving malformed regions.

Note that it is possible to have regions that don't meet the criteria above, but they will be clearly flagged as potentially problematic.

For example, regions of interest have no validation criteria, so a user could create a "bad" region of interest and then activate training or inference behaviors for it.

Beyond that, legacy regions and regions imported via the geojson import tool will be allowed without any validation. The warnings will still show, however, when trying to use them.

We are also changing the yearly plan credit policy to give yearly subscribers more flexibility.

Changelog​

Changes

Fixes

We also have added support for delta layer downloads and improved delta layer UX in general!

Changelog​

New Features and Improvements

Fixes

The main highlights include Access Control, new SmartSeg models, and Generic Image File Support. But there's much more to see, so be sure to check out the full changelog below!

Changelog​

Major New Features

New Features and Improvements

Changes

Fixes