sahi
Framework agnostic sliced/tiled inference + interactive ui + error analysis plots
What is sahi?
Framework agnostic sliced/tiled inference + interactive ui + error analysis plots
Overview of sahi
What it is
SAHI (Slicing Aided Hyper Inference) is a lightweight vision library for performing large scale object detection and instance segmentation. It enables sliced inference for detecting small objects in large images and supports various popular detection models.
Key features
- Sliced/standard video and image prediction using Ultralytics, MMDetection, Hugging Face, TorchVision, YOLOv5, and Roboflow models.
- COCO dataset slicing, evaluation, error analysis, and conversion to Ultralytics format.
- Interactive result visualization and inspection via FiftyOne integration.
- CLI commands for prediction, evaluation, and analysis.
- Support for video inference.
- Error analysis plots and evaluation metrics (COCO AP/AR).
Use cases
Detecting small objects in large images (e.g., satellite imagery, aerial photos, medical scans). Performing large-scale object detection and instance segmentation in research and production. Evaluating and analyzing model predictions on COCO datasets. Visualizing and inspecting detection results interactively.
Getting started
bashpip install sahiFor detailed installation instructions (including PyTorch, TorchVision, and detection framework setup), see the repository README.
License and pricing
This project is licensed under the MIT license. It is open source and free to use.
Based on the project's public README and repository metadata.