sahi-inference - Object detection on large images using SAHI tiled inference
Performs object detection on large images by splitting them into overlapping tiles and merging predictions.
Tags
Updated: 2026-09-21Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Load detection models with AutoDetectionModel
- Execute sliced inference on large images
- Configure slice dimensions and overlap ratios
- Postprocess tile predictions using NMS
- Generate SAHI Python inference code
- Optimize SAHI inference pipelines
- Explain SAHI slicing and merging
- Design tiled inference learning exercises
Inputs
- Large images or drone orthomosaics
- Object detection model weights
- Slice dimensions and overlap ratios
- Postprocessing options and thresholds
Outputs
- Full-image bounding box detection results
- Generated SAHI Python code
- Pipeline performance optimization recommendations
- Tiled inference exercises and solutions
Requirements
- Python package sahi
- Supported backend packages such as ultralytics or torchvision
- CUDA or MPS device