Labeling geospatial imagery
Aerial images are usually high-resolution images with large coverage. Their sizes can reach hundreds of megabytes (or more). Tiling is often necessary to navigate all the areas of the image and zoom in on specific areas to annotate objects.
For annotating geospatial/satellite imagery, Kili currently supports GeoTIFF files (.tif files with geospatial metadata). The standard tiff format (.tif/.tiff) is not natively supported by Kili.
Multi-band imagery support
We only support geospatial files with the following bands :
- 1 band (gray)
- 2 bands (gray + alpha)
- 3 bands (RGB)
- 4 bands (RGB + alpha)
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_translate
utility 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)
When uploaded to Kili, all images larger than 30MB are automatically tiled, for better performance. This means that, technically, labeling GeoTIFF images is not different from labeling standard images and you can use all the standard tools that you’d normally use with Image projects. The only differences are:
-
When GeoTIFF files are detected, Kili switches off the asset rotation feature, to preserve the original geo-coordinates.
-
When annotating multi-spectral GeoTIFF images, you can toggle their 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. For more information, refer to Adding a multi-layer image.
- 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.
- When annotating GeoTIFF images, 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.
- When exported, GeoTIFF labels are saved together with their corresponding geographic coordinates sourced from the image geospatial metadata.
- In exported annotations,
x
stands for longitude andy
for latitude.- For details on what TIFF types are supported by Kili, refer to our list of supported TIFF raster drivers.
Additionally, in Image projects with GeoTIFF assets, 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.
This feature is available through Kili SDK. For more information on how to use minZoom
and maxZoom
values, refer to Kili SDK documentation.
- 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.
Updated 5 months ago