Use the model - Pixel Classification

Geospatial Vision Language Model (GeoVLM) beta is available through Esri's Early Adopter Community. Please sign up here to request access to the model. You can use GeoVLM for pixel classification in the Classify Pixels Using Deep Learning tool available in the Deep Learning toolset of the Image Analyst toolbox in ArcGIS Pro.

1. Download the Geospatial Vision Language Model and add any imagery layer in ArcGIS Pro.

   

2. Click the Analysis tab and browse to Tools.

   

3. In the Geoprocessing pane, click the Toolboxes tab and expand Image Analyst Tools. Select the Classify Pixels Using Deep Learning tool under the Deep Learning toolset.

   

4. On the Parameters tab, set the variables as follows:
   1. For Input Raster, select the image.
   2. For Output Raster Datset, set the output raster that will contain the classification results as a binary raster representing the prompted object class and the background class.
   3. For Model Definition, select the model .dlpk file.
   4. Optionally, for Arguments, modify the values of the arguments to optimize performance.
      • Text Prompt—Text that describes the object to be detected. For example, buildings, swimming pools, impervious surfaces, and so on.

        Note:

        Multiple objects cannot be prompted in a single run.

      • Threshold—Pixels with a confidence score higher than this threshold will be assigned to the object class. The default value is 0.5. The allowed values range from 0 to 1.0.
      • Batch Size—The number of image tiles processed in each step of the model inference. Increase the value to process more tiles together to have smoother transitions when merging tile outputs, reducing visible seams and boundary artifacts. The default value is 4.
      • Predict Background—Specifies whether the background class will be classified. If true, the background class is also classified. The default value is false.
   5. Check Use pixel space if you want inferencing to be performed in pixel space and the output to be transformed back to map space.

      This option is useful when using oblique imagery or street-view imagery, which may cause the features to become distorted using map space.

   
5. On the Environments tab, set the variables as follows:
   1. For Processing Extent, define the geographic area of interest for inferencing.

      The whole image extent is used if undefined.

   2. Set the Cell Size value in meters in such a way that maximizes the visibility of the objects of interest in the inferencing tiles. Consider a larger cell size for detecting larger objects and a smaller cell size for detecting smaller objects. For example, set the cell size to 0.2m or 0.3m for buildings, 0.5m or 1m for trees and roads, 5m or 10m for huge lakes and clouds. For further information regarding cell size, refer to the Cell size of raster data resource.
   3. For Processor Type, select GPU. If multiple GPUs are available, set GPU ID to the GPU to be used.
   
6. Click Run.

   As soon as processing finishes, the output classified raster is added to the map.