Introduction

Geospatial imagery annotation involves working with specialized image data captured from satellite, aerial, or drone sources. These images differ significantly from standard photographs in several ways, requiring tailored approaches and tools for effective labeling. Below are some key specificities of geospatial imagery:

• High Resolution and Large File Sizes

Geospatial images are typically extremely large, with high ground resolutions where each pixel corresponds to a specific area of the Earth's surface. These files can be hundreds of megabytes or more in size, making smooth loading, navigation, and zooming capabilities essential for annotators.

• Geo-coordinates Metadata

Unlike ordinary images, geospatial imagery includes geographic metadata, where each pixel is tied to precise geo-coordinates. This enables annotators to work with spatial accuracy, ensuring annotations align with real-world locations.

• Multispectral Data

Geospatial images often go beyond the standard RGB format, including multiple spectral bands such as infrared, thermal, or others. These bands provide additional data, essential for tasks like vegetation analysis, water body detection, or urban planning.

• Support for Different Projections

Geospatial imagery is represented using various coordinate reference systems, often specified by EPSG codes. These projections affect how the Earth's curved surface is translated into a flat image, and annotation tools must be compatible with diverse projection systems.

The specificities of geospatial imagery demand purpose-built capabilities to ensure efficient and accurate annotation workflows.

---

Uploading Geospatial images

Our platform supports the seamless upload of geospatial data with the following specifications and options:

Supported File Types and Sizes

• File Formats: GeoTiff images, including .tif, .jp2, and .ntf files.
• Maximum File Size: Up to 1 GB per file. For files larger than 30 MB, Kili preprocesses the image to generate tiles based on the specified zoom levels. These tiles are then used to optimize navigation in the labeling interface.

Multi-Band imagery Support

We support a variety of band configurations to accommodate different use cases: 1-band (gray), 2-band (gray + alpha), 3-band (RGB), and 4-band (RGB + alpha) configurations.

📘

Multi-band image upload

If you upload an asset with more bands than the ones mentioned here above, we will import the first 3 non-alpha bands as RGB bands. If you have files with more bands and if this behavior is not the one you expect, you will need to preprocess the image before uploading it, by selecting 1 (for grayscale) or 3 bands (for RGB) among all the bands to create a new image with these.

For this you can use the gdal_translateutility from GDAL as follows:

• gdal_translate -b $BAND_NUMBER input.tif output.tif (for 1 band)
• gdal_translate -b $BAND_NUMBER_1 -b $BAND_NUMBER_2 -b $BAND_NUMBER_3 input.tif output.tif (for 3 bands)

Supported Projection Systems

• EPSG4326 (WGS 84).
• EPSG3857 (Web Mercator).

Importing Geospatial Data

Via the Import Menu

• You can upload single-layer assets using the image's existing projection system.

Via the SDK

You can:

• Modify the image’s projection system: you can convert and upload images in either of the supported projections, EPSG4326 or EPSG3857. Ensure consistency between layers to maintain spatial accuracy.
• Upload all layers using the same projection system to ensure alignment and compatibility. For more information, refer to Adding a multi-layer image.
• You can specify the minZoom and maxZoom values. The value range for both is 1-20 (level 1 being the lowest-detail: basically just an image of our planet, and level 20 being the maximum zoom for the asset).
  The minZoom parameter defines the zoom level that users are not allowed to zoom out from. The maxZoom value affects asset generation: the higher the value, the greater the level of details and the size of the asset. For example, if you set minZoom to 15 and maxZoom to 20, the asset will be uploaded with the maximum level of details, will take longer to load and process, and will take up a lot of space. When opened, the example asset will show as zoomed in to a level sufficient to show specific city districts and won’t allow users to zoom out.
• Add a base map: to provide geographical context, you can integrate a base map using any available tile server. Examples include Open Street Map and Google maps

For detailed instructions on SDK usage, refer to the tutorial here.

---

Annotating images

Our platform provides intuitive tools and features to facilitate efficient annotation of geospatial imagery.

Panning

Click and drag the image to move to a different area. As you pan, tiles will load dynamically to ensure smooth navigation and maintain high performance, even for large images.

Zooming

• Press Space + Draw a bounding box to zoom in on a specific area.
• Adjust zoom levels directly from the header or by using the + and - keys
• Press Shift+ 1 to go back to the initial zoom level

Rotating the image

You can use the rotation button to adjust the orientation of geospatial images. With each click, the image rotates by 10 degrees. Alternatively, you can enter the desired rotation angle directly into the input field for more precise adjustments. If you need to reset the rotation to its original orientation, simply double-click the rotation button.

Working with multiple layers

You can toggle layers, so you can visualize the same spot from different spectral views.This is useful in situations when you want to get the historical context of the same scene at different points in time or you need to visualize different spectral views, each one corresponding to a specific combination of the original image’s bands to annotate with extra accuracy. You can also use a slippy map (such as OpenStreetMap) as a base layer, for precise image location reference when making annotations.

Copying Coordinates

You can quickly copy the geo coordinates of any point. To do that, hover over it, click the right mouse button, and then select Copy GPS Coordinates. Alternatively, use the Alt/Option + C keyboard shortcut. The coordinates are copied to clipboard.

Measuring Distances

You can measure distance between any two points. The distance is provided in meters and feet. This is helpful in situations when an object’s length is a decisive factor for assigning it a specific class.

Creating Annotations

Labeling geospatial images is not different from labeling standard images and you can use all the standard tools that you’d normally use with Image projects, including Interactive Segmentation with SAM2.

---

Exporting labels

When working with geospatial images, exported labels include the associated geo-coordinates, ensuring spatial accuracy and usability across geospatial applications.

Supported Export Formats

• Kili Format: A versatile format for integration with Kili workflows and further processing.
• GeoJSON Format: Widely used for geospatial data, compatible with various GIS tools and applications.

For detailed steps on how to export labels, refer to the export documentation here.

📘

Exported Geographic coordinates

• In exported annotations, x stands for longitude and y for latitude.

---

Importing labels

Kili Format

Directly import labels saved in Kili format.

GeoJSON Labels

Convert GeoJSON labels into Kili format using our utility functions. You can find more information and the relevant code examples in the utils documentation. For a detailed guide on importing and converting labels, refer to this tutorial.