Birdi’s Map AI Detect tool lets you automatically identify objects directly on your orthophotos, saving hours of manual annotation time. This feature is currently in beta and available on all Birdi plans.
If you're looking to detect objects in uploaded images or videos, see the Media AI Detect article instead:
1. Getting started
Open your map project in Birdi.
Select the New → AI Detect button.
Choose the Map option.
This will open the AI Detect panel where you can configure how the detection runs.
2. Choose the orthophoto
From the dropdown list, select the orthophoto you want the AI to analyze.
This will be the base layer used for all detections.
3. Enter what you want to detect
In the prompt field, type the name of the object you want the AI to find, for example:
“cars”
“trees”
“roofs”
You can enter a single word or a short phrase.
4. Choose annotation type: Point or polygon
You can choose how you want the detected outputs to be represented on the map:
Point
Best for counting objects
Simple, lightweight, and easy to review
Polygon
Best for area-based objects
Useful for measurement, volumetric calculations, and stockpile conversion
5. Understanding the AI settings (Recommended defaults included)
Birdi automatically chooses the best recommended settings for your map using an LLM that analyses:
the size of your orthophoto
the resolution
the type of object you want to detect
These settings work well for most detections, but you can adjust them under Advanced Settings if needed.
Advanced settings explained
Threshold Confidence
Higher confidence = more accurate results, but may take longer to process.Tile Size
Defines how much of the map the AI analyzes at once (larger tiles may speed things up but can reduce fine-detail detection).Tile Overlap
Controls how much tiles overlap. More overlap improves accuracy along tile edges.
6. Run the detection
Select Detect Objects to start the process.
While the AI Detect tool runs:
A map notification will appear showing progress
A new folder in the Layers panel will be created
Annotations will begin populating as they are processed
You can continue working while this runs.
7. You can close the panel anytime
You can safely close:
the AI Detect panel
the notification
or even the map
The detection continues running on the server, and results will be delivered back into your map automatically.
8. Completion email
When the detection is finished, you will receive an email notification.
Processing time can vary significantly depending on:
the object type
the size of the orthophoto
the complexity of the scene
9. Review your AI annotations
Once complete, your detections will be visible in the Layers panel.
You can now:
Review the annotations created
Refine them by deleting, editing or styling > How to customize and style your annotations
Take measurements using the table view > How to access and use Table View for your 2D annotations
Convert polygons over stockpile to volumetric annotations
Export them in different formats (CSV, Geojson,KML or shape file)
Include them in assessments and reporting > How to use the map canvas export tool
10. Tips & Notes for map AI Detect
Image quality matters
The accuracy of AI detection is highly dependent on the quality and resolution of your orthophoto. Lower-quality imagery, heavy shadows, or blurred areas may impact results.
If you continue to experience issues, please contact our team via the in platform chat.
Edge-of-Map Detections
Objects located near the edge of the orthophoto can sometimes be misidentified or partially detected.
If this happens, select the annotation in the layers panel, right-click, and choose Delete.
Over-Extended Polygons
Sometimes a polygon may extend beyond the intended object due to:
shadows
shape distortion
ground warping
overlapping materials
You can correct these by:
Selecting the annotation
Pressing Enter to enter edit mode
Adjusting the vertices as needed, if you click on the points directly it will delete it and simplify the polygon
If the annotation is too inaccurate, delete it and manually draw a new one.
Trying different object terms
Detection performance can vary depending on:
the wording used
the type of object
the complexity of the scene
If you’re not getting good results, try using common, simple terms like “car,” “truck,” “tree,” or “stockpile.”
Review before using data
Always review the AI annotations before using them for:
measurements
reporting
stockpile calculations
assessments
Minor corrections may be needed, especially for complex sites.
Run times vary
Processing time depends on:
orthophoto size
number of tiles
object complexity
confidence thresholds
Larger or more detailed maps may take longer to complete.
Server-side processing continues
You can safely close the map or browser — the detection continues on the server, and your results will return to your project automatically.
Multiple orthophotos
If your project contains multiple orthophotos, ensure you’re selecting the correct dataset for detection. We can only process one orthophoto at a time in map.
Layer organisation
Detected objects are stored in their own automatically created folder — renaming the folder can help keep maps organised if running multiple detections.
Advanced settings changes
Small adjustments to confidence, tile size, or tile overlap can significantly influence outcomes. If unsure, start with the recommended defaults.
Mixed object scenes
If your map contains densely packed or overlapping objects, running detection in smaller sections or with adjusted settings can improve clarity.
If in doubt, reach out
AI detection won’t always be perfect, especially in noisy, shadowed, or cluttered environments. If you’re unsure or need help tuning your settings, our team is here to support you.