Introducing Results Annotation
Our customers leverage GO to analyze vast amounts of satellite imagery, allowing them to continuously monitor areas of interest at scale, and making analysts' time more efficient by automatically identifying trends and anomalies.
At the same time, we recognize that AI and computer vision will never be 100% perfect, and that many customers need an even higher level of human accuracy, particularly in mission-critical environments.
To address this, we are releasing Results Annotation as a beta feature, which will allow users to "annotate" object detection results by adding missed detections (false negatives) and removing erroneous detections (false positives) from results.
Benefits to Customers
- Continue to leverage the scale and efficiency benefits of GO - automated analysis of imagery at scale with computer vision (CV)
- Augment results with annotation to go from CV-level accuracy to human-level accuracy, for mission-critical situations and reports
- Over time, benefit from feedback from subject matter experts through identification of failure modes, that can inform our algorithm improvement process (NB: customer annotations may be used only with your explicit consent, and following a vigorous QA process)
This feature is a beta release - we would love to hear your feedback & suggestions!
*In the current beta phase, we are focused on improving the speed and user-friendliness of the annotation feature, as well as iron out any bugs we identify in this process.
Release Notes (14 April 2021)
- This release enables annotation for single-class object detection algorithms (cars, trucks, and ships)
- Annotation for multi-class algorithms will follow shortly (UPDATE 9 June: this is now available)
- Annotation is only available for newer projects run after 1 April 2021
- Annotated results are available via UI and API. We will look to make them available via download in future. (UPDATE 11 May: this is now available via download)
- When turning on annotation while heavily zoomed-in on the map, object detections may disappear from the map. Zoom out and then back in again to load & view the objects. (UPDATE 29 April: this has been fixed)