Skip to main content

FlyPix Release: v3.7.0

ยท 5 min read

Version 3.7.0 introduces external imagery import, allowing you to reference public Cloud-Optimized GeoTIFFs, connect to WMTS services, or create rasterless files scoped to any geographic extent โ€” all without uploading a file. This release also deepens Geosense capabilities with a Find Similar integration and session contextualization warnings, adds Shapefile and GeoPackage upload support, and improves how the app handles annotations with geometry issues.

Changelogโ€‹

Major New Features

๐Ÿ’ฅ
Import External Imagery (WMTS, Public COG, Rasterless)

You can now add imagery to a project without uploading a file. Three new source types are available from the Add Imagery dialog:

  • Public COG โ€” reference a Cloud-Optimized GeoTIFF hosted at any public URL.
  • WMTS โ€” connect to a Web Map Tile Service endpoint and select a layer.
  • Rasterless โ€” create a synthetic raster scoped to a geographic extent, rendered using a base map provider.

You can also drag and drop a .geojson file onto the imagery upload area to automatically derive a Rasterless file from the file's bounding box โ€” no upload required.

See the Import External Imagery guide for full details.

New Features and Improvements

โœจ
Geosense Find Similar Integration

You can now use our flagship Find Similar tool by asking Geosense to find a certain type of annotation. This will use available samples in your context and a textual prompt to find similar annotations in your viewport.

โœจ
Copy Geosense Results to New Layer

You can now copy annotations generated by Geosense directly into a new vector layer with a single click. This is available as an action in the inspector when you have Geosense-generated annotations selected.

This creates a new layer with the same class as the source annotations, and copies all selected features into it. This is useful for quickly separating out Geosense results for further editing or analysis without modifying the original layer.

โœจ
Geosense Session Contextualization

You can also see which sessions are miscontextualized (i.e. were not created in the current file), to avoid mixing up sessions from different files and thus have a more consistent experience when working with Geosense. This also mitigates issues we have seen with Geosense annotation references not loading when in the wrong context.

All prior sessions have been migrated to be contextualized in the project. So this feature will not flag them, but they still work as expected. Only newly created sessions will be flagged as miscontextualized.

โœจ
Geosense Chat Refinements

We have improved the visual design of some Geosense chat elements. Notably, the "thinking" chat bubble is now a lot more lively and features an animation to keep Geosense more dynamic. We also show synthetic responses to messages like Questions or the new "Find Similar" Geosense tool in a visually distinct style from your own messages.

โœจ
Improved Bad Geometry Handling

We were inconsistently handling bad geometries (e.g. those with self-intersections) in the app, leading to cases where some annotations would suddenly disappear or fail to load.

We now detect these and display them with a warning style, showing detailed issues in the inspector to help you fix them. Note that these annotations with issues will typically be ignored by certain operations like training or change detection, so fix them as soon as possible!

โœจ
Shapefile & GeoPackage Upload Support

We have introduced support for uploading Shapefiles and GeoPackages directly into the app. Right now, support for these formats is fairly limited, and the format we require is very strict. For details on these requirements, consult the API Specification, under Vectors > Upload Gpkg / Upload Shp.

โœจ
Thumbnail Transparency

Raster data with transparency or NODATA values will now result in thumbnails with transparency instead of a black background.

Changes

๐Ÿ”„
Geosense Default Contexts

We are now including context by default on Geosense messages. This hopefully speeds up iteration by reducing the number of questions coming from the agent about missing context. You might need to deactivate some contexts if you want the agent to actively ignore them.

๐Ÿ”„
Advanced Training Cost Scaling

The cost of training a model is now scaled based on the number of epochs used. So you can benefit from lower costs when training with fewer epochs for faster iteration. Conversely, longer trainings with more than the default epochs will have an increased cost.

Fixes

๐Ÿ”ง
Fixed a Geosense session loading issue causing annotations not to resolve until you sent a message.
๐Ÿ”ง
Fixed a minor caching issue causing model renaming in the details page not to be reflected without a page refresh.