# obss/sahi

Framework agnostic sliced/tiled inference + interactive ui + error analysis plots

Repository: https://github.com/obss/sahi
Canonical: https://ross.abutalabs.com/products/sahi
Homepage: https://obss.github.io/sahi/
Language: Python
License: MIT
License Family: permissive
Topics: object-detection, instance-segmentation, computer-vision, small-object-detection, large-image, pytorch, python, coco, deep-learning, machine-learning, remote-sensing, huggingface, fiftyone, satellite, tiling, explainable-ai, oriented-object-detection, hacktoberfest, yolo26, open-vocabulary-detection
Last push: 2026-08-22T20:39:30+00:00

## Health v2 (maintenance only)
Score: 99/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 98, longevity 100
- inputs: {"age_days": 2041, "days_push": 11, "days_rel": 17, "gap_med": 7.5, "n_releases_24m": 25}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5473, forks 780 (observed 2026-08-28T04:09:19.624008+00:00)

## What it is
SAHI (Slicing Aided Hyper Inference) is a Python vision library for detecting small objects in large images via sliced/tiled inference, working with any object detection or instance segmentation model without retraining. It includes an interactive UI, error analysis plots, COCO dataset utilities, and CLI commands.

## Use cases
- detect small objects in large aerial or satellite images
- improve object detection accuracy on high-resolution imagery
- run sliced inference with YOLOv8 or HuggingFace detection models
- tile large images for object detection without retraining the model
- analyze detection errors on COCO-format datasets
- detect tiny objects in drone or surveillance footage

## When to choose
- you need better recall on small objects in large images
- you want a model-agnostic slicing pipeline supporting YOLO, MMDetection, Detectron2, or Transformers
- you work with aerial, satellite, or drone imagery
- you need COCO dataset slicing, splitting, and filtering utilities

## When to avoid
- your images are small and objects are large - standard inference suffices
- you need real-time low-latency detection, since tiling adds overhead
- your task is classification or keypoint detection rather than detection/segmentation

## Facets
- artifact type: library
- maturity: active
- function: computer-vision, image-processing, machine-learning, cli, data-visualization
- domain: computer-vision, deep-learning, machine-learning
- platform: python, cli, cross-platform
- tags: object-detection, instance-segmentation, sliced-inference, small-object-detection, tiling, coco, yolo, huggingface, satellite-imagery, aerial-imagery, remote-sensing

## Member repositories
- obss/sahi (main) score 99

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:19.624008+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:56:43.422633+00:00, confidence not recorded.
  - readme: https://github.com/obss/sahi (fetched 2026-08-28T04:09:19.624008+00:00, sha 0d0dfb03038d)
  - homepage: https://obss.github.io/sahi/ (fetched 2026-08-29T08:51:45.067725+00:00, sha b979e9b37437)
  - registry_pypi: https://pypi.org/pypi/sahi/json (fetched 2026-08-29T08:51:45.077006+00:00, sha ba5216593501)
- Data as of 2026-08-30T08:39:29.467469+00:00.