Semantic Feature Extraction

Function Overview

Semantic Feature Extraction extracts semantic vector features from raster images according to user-defined prompts. The tool supports common feature categories such as roads, buildings, and water bodies, and can optionally regularize, simplify, and smooth the vector results.

Usage

Click Photo > Semantic Feature Extraction to open the tool.

Semantic Feature Extraction

Parameter Settings

  • Input Image: Input image file. Supported formats include *.tif, *.tiff, and *.img.

  • Output Vector: Output vector file. Supported formats include *.gpkg and *.shp.

  • Prompt: Semantic categories to be extracted. Multiple categories should be separated with commas. The default categories are road, building, and water.

  • Confidence Threshold: Confidence threshold for keeping extraction results. The valid range is 0 to 1, and the default value is 0.3.

  • Image Tile Size: Tile size used for inference. Available values are 512, 1024, 2048, 4096, and 8192. The default value is 1024.

  • Inference Scale: Scale factor used during inference. Available values are 1, 2, 4, and 8. The default value is 1.

  • Overlap Buffer: Overlap width between adjacent tiles. The valid range is 0 to 1024, and the default value is 102.

  • Minimum Area: Minimum polygon area kept in the output result.

  • Enable Regularization: Regularize extracted vector features.

  • Regularization Layer: Layer to be regularized. The default layer is building.

  • Regularization Angle Threshold: Angle threshold used during regularization. The valid range is 0 to 90, and the default value is 20.

  • Regularization Distance Threshold: Distance threshold used during regularization. The default value is 2.

  • Enable Simplification: Simplify extracted vector features.

  • Simplification Layer: Layer to be simplified.

  • Simplification Tolerance: Tolerance used for vector simplification.

  • Enable Smoothing: Smooth extracted vector features.

  • Smoothing Layer: Layer to be smoothed. The default layers are road, water, vegetation, grass, and shrub.

Notes

  • Currently, only English prompts are supported.

  • For smooth operation, an NVIDIA GPU with Compute Capability 8.0 or higher is recommended. This corresponds to NVIDIA consumer GPUs in the RTX 30 series or later. GPUs below Compute Capability 8.0 may experience stuttering.

Output

The tool outputs a semantic feature vector file.

results matching ""

    No results matching ""