The View button on the toolbar allows you to control various options which influence how your detections are displayed. You can switch between viewing the detections as shapes and as the center points of those detections. You can also toggle showing or hiding all detections from the image here.
Results
Detection area
In order to define specific regions within your image for detection, you can assign what we refer to as “Detection Areas.” These areas are polygons that delineate the regions in the image where the detector will be applied.
It’s important to note that this feature is not intended for extremely precise image clipping, but rather provides a way to broadly indicate the regions the detector should focus on. This is because, for performance reasons, we apply some geometry simplification when uploading them to the platform.
To set detection areas for an image, click the ‘…’ menu next to your image and open the ‘Info’ popup. You will find a ‘Manage detection areas’ line with a manage button that you can click.
This will open the image view screen, with already opened the detection areas tool and the detection areas panel.
Detection Areas Panel
This panel allows you to control whether detection areas are displayed (the checkbox), influence how they are shown (either a pin at the center of the shape, or the full shape drawn on the image), and navigate between them with the <
and >
buttons. Navigating between the detection areas will automatically zoom the map to the location of the detection area.
Clicking on the (…) menu will reveal options to upload or download your detection areas as GeoJSON, and provide the option to delete all detection areas and revert to the default (the full footprint of the image) for detection.
Detection Areas Tool
From the Detection Areas Tool, you can select one of three options:
Edit Detection area
Draw a new Detection Area
Upload Detection Areas
There are three main requirements for this file:
- Your geometries’ coordinates should be projected in EPSG:4326.
- All geometries should be valid. You can check this in QGIS with the Vector → Geometry Tools → Check Validity tool.
- Your geometry should cover parts of the image on which it is applied.
Here’s an example of a valid Detection Area file:
{
"type": "FeatureCollection",
"features": [
{
"type": "Feature",
"properties": {},
"geometry": {
"type": "Polygon",
"coordinates": [
[
[
6.573850214481354,
46.52660109443889
],
[
6.574483215808868,
46.52660109443889
],
[
6.574483215808868,
46.52700708253873
],
[
6.573850214481354,
46.52700708253873
],
[
6.573850214481354,
46.52660109443889
]
]
]
}
}
]
}
Generate results
Detecting objects is quick and easy using Picterra:
Manage results
The Image View
Image Info
The image info gives you context for your image, showing you the image name and the project it is stored in.
Bands and Band Settings (multispectral images only)
If your image is multispectral, the bands dropdown allows you to pick which band visualization setting you want to display. The < and > buttons allow you to cycle through the available band visualizations, and the dot menu … gives options to edit the band settings.
Image Toolbar
View
Split view
Selection
Drawing tool
Detection areas
Magic wand
Markers
View
Split view
If your image is multispectral and you have more than one band visualization defined (more info), you can view two band visualizations side-by-side. Detections display and drawing will only be enabled on the primary image (left side), but a duplicate cursor will track your actions on the secondary image (right side).
Selection
This default tool allows you to:
- Click and drag to move the map
- Left Click to Select a Single Polygon for Editing
- Shift-click and drag to draw a selection box to select multiple polygons
Click and Drag to move the map
If you click and hold, and drag your mouse you can scroll around the image.
Select a Single Polygon for Editing
Selecting a single detection allows you to move and edit the polygon. A number of actions are available to you including:
- Rotating and resizing the shape
- Click and drag the point on the dotted circle to rotate or resize the shape
- You can click the Fix Size Rotation button to lock the current shape size and restrict this to rotation only
- Move the shape
- Click inside the shape and drag it to move it
- Use the shape as a template and clone copies of it
- Click the Clone button. Your cursor will change to an outline of the shape. Left click to place copies.
- Cut holes in your shape
- Click the Cut Inside button, you will now be able to draw another shape inside your shape to cut a hole. Double-click to finish drawing.
- Delete the shape
- Press the
Delete
orBackspace
key to remove the shape
- Press the
At any point you can discard or save your edits with the Discard and Save buttons respectively.
Selecting Multiple Polygons
To select multiple polygons, you can hold down your Shift key and click and drag to draw a selection box around the polygons you wish to select.
To delete the polygons you have selected, press Delete or Backspace.
Drawing tool
Select this tool to draw a new polygon outline in your currently selected detections layer. Left click on the image to start drawing, click to place points around the edge of your shape and double click to finish drawing.
Detection areas
From the Detection Areas Tool, you can select one of three options
Edit Detection Areas
This tool allows you to select existing detection areas and modify their shape, move them and delete them.
Draw a new Detection Area
This tool allows you to draw a new shape to create a detection area. Detection areas determine where detections will be made on your image when running your detector. Areas outside of detection areas will not be considered by the detector. Left click on the image to start drawing an area, click again to place points and double-click to finish drawing.
You can upload a geojson file which represents detection areas. You can drag and drop a file onto the dialog, or click to select one from your computer. We currently support geojson Polygon and Multipolygon feature types.
Magic wand
The Magic Wand tool allows you to quickly create outlines out of complex objects or regions in just a few clicks. This is leveraging the “Segment Anything” model by Meta/Facebook.
Markers
The markers tool allows you to leave comments at a specific location on your image. With this tool selected click anywhere on the map to leave a new comment, use the @ symbol to mention colleagues or leave yourself a reminder. You can read more about our collaboration features here.
Panels (Accordion)
Detection area
Layers
Markers
Attributes
Detection area
A detection area/s is the region within the detector runs and identifies results.
Layers
Running a detector outputs a new layer of detections. You can think of a layer as a collection of shapes (polygons).
The layers panel allows you to control the visibility of each individual layer, assign colors to layers, export layers, and to create new layers either manually or via upload.
Exporting Layers
To export a single layer, click the dot … menu in-line with a layer, and choose Export. We support exporting to geojson, shapefile (.shp), csv or kml formats.
To export multiple layers, ensure that the checkboxes associated with the layers you wish to export are selected and then click the dot … menu at the top of the layers panel and choose Export Selected. Your selected layers will be downloaded in your chosen format inside a .zip file.
The Layer Attributes button
The attributes button in the layer panel allows you to add or remove attributes associated with your layer. Attributes are metadata associated with detections.
Markers
The markers panel allows you to navigate between markers using the <
and >
buttons, filter markers to only those that mention you, and to toggle display of markers on and off using the checkbox.
Attributes
An attribute is additional information or metadata that users can associate with their detections to provide context or extra details about the detections.
Attributes
What are attributes?
An attribute is additional information or metadata that users can associate with their detections. These attributes provide context or extra details about the detections, allowing for better organization and analysis of the data. Attributes help to categorize, label, or describe the detections more comprehensively, enhancing the overall understanding of the data.
Picterra supports three types of attributes:
Attribute Type | Default Value | Input Type |
Text | Text Input | |
True / False | False | Checkbox |
Number | 0 | Number Input |
An attribute is associated with the layer – when an attribute is added to a layer, it applies to all detections within that layer with the default value. Removing an attribute removes it from all detections in the layer.
WARNING – Be careful when removing attributes, as this action will erase all values assigned to attributes across all detections within the selected layer.
The attributes panel displays the attributes associated with the currently selected polygon. For adding further attributes, please consult the Layers panel documentation provided above.
Example of attributes you can find / or define in Picterra:
- Confidence
- Height
- Area
Attributes visualization per layer
In addition to viewing the attribute per polygon, you can also visualize them on a per-layer basis using a color gradient. Upon activating this function, polygons with the lowest attribute values in the range will display as black, while those with the highest values will adopt the color of the detection layer. This gradient visualization is compatible with any numerical attribute, such as height or area. In the near future, we plan to introduce more attributes automatically associated with your detection results. Furthermore, we aim to include support for numerical properties imported onto your own vector layers.
If you have a large number of polygons over a large area, you can switch them to markers mode and then markers will be colored using the attribute gradient – see below an example with confidence gradient.
If you have a large number of polygons over a large area, you can switch them to markers mode and then markers will be colored using the attribute gradient – see below an example with confidence gradient.
This can give you a good idea of which regions in the image your detector is less sure of, which can help guide your decisions on whether or not to add more training data in those regions, or to draw a detection area to exclude them. You can read more about detections confidence here.
Types of attributes
Detection confidence
In Picterra, each detection polygon is associated with a confidence attribute, representing the detector’s certainty in its prediction. This attribute indicates the level of confidence the detector has in its output, akin to saying, “I’m 90% sure that I outputted the right thing here.”
How is confidence computed?
Confidence is computed using the direct output of the deep learning model, loosely related to how uncertainty maps are calculated in accuracy areas. Think of it as 1 minus the uncertainty value averaged across a detection, focusing on the inner 50% of the detection’s area to mitigate the influence of high uncertainty at the outer edges.
While confidence computation works in both count mode and segmentation mode, the results may be more interpretable in count mode. In segmentation mode, a very large polygon will yield only one confidence value for the entire detection.
Attributes visualization per layer
In addition to viewing the attribute per polygon, you can also visualize them on a per-layer basis using a color gradient. Upon activating this function, polygons with the lowest attribute values in the range will display as black, while those with the highest values will adopt the color of the detection layer. This gradient visualization is compatible with any numerical attribute, such as height or area. In the near future, we plan to introduce more attributes automatically associated with your detection results. Furthermore, we aim to include support for numerical properties imported onto your own vector layers.
Utilizing confidence
Analyzing individual detections and their confidence values provides direct and detailed insights into the detector’s strengths and weaknesses. Zooming out to view the center of the raster, colored by confidence, offers an overview of regions posing challenges for the detector.
Note: You may encounter results such as occasional detections that are clearly correct with low confidence, or incorrect detections with high confidence. This could indicate confusion in the detection process and highlight potential weaknesses. Generally, true positive results will have higher confidence than false positive ones, allowing you to set a threshold for ignoring certain detections and improving overall accuracy.
Future enhancements
Future updates will introduce tools for bulk filtering detections based on confidence value ranges. This functionality will enable filtering out lower confidence polygons, allowing for a precision-focused output with an overall accuracy boost.