QPredict - Model Inference
QPredict is the GeoAI suite extension dedicated to running artificial intelligence models directly within the QGIS environment. It allows models trained with QModel Trainer to be applied to georeferenced raster data, producing raster or vector layers that can be used immediately within a GIS project.
Objectives of QPredict
- Enable inference of GeoAI models trained within the GeoAI suite
- Integrate artificial intelligence workflows directly into QGIS projects
- Produce georeferenced results ready for analysis or cartographic use
- Ensure reproducibility and traceability of inference processes
Supported model types
QPredict supports several families of deep learning models commonly used in GeoAI, including PyTorch-based segmentation and classification models, Mask R-CNN architectures, and the latest YOLO models for object detection, segmentation, and oriented objects. The type of output generated depends on the model used.
General workflow
- Loading a trained model using the GeoAI suite dedicated format
- Selecting a raster layer from the QGIS project as input data
- Defining the spatial extent to be processed
- Running inference with progress monitoring and logging
- Automatically generating output layers in the project and on disk
Design philosophy
- Guided and consistent interface aligned with other GeoAI suite extensions
- Clear separation between input data, processing parameters, and outputs
- Detailed logging for monitoring and diagnostic purposes
- Results immediately visible and usable within QGIS
QPredict serves as the operational link between GeoAI model training and their practical application on geospatial data, within a controlled GIS environment.