# mahyarnajibi/SNIPER

SNIPER / AutoFocus is an efficient multi-scale object detection training / inference algorithm

Repository: https://github.com/mahyarnajibi/SNIPER
Canonical: https://ross.abutalabs.com/products/sniper
Language: Python
License: NOASSERTION
License Family: other
Last push: 2021-08-22T01:17:53+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 3154, "days_push": 1838, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2690, forks 435 (observed 2026-08-28T04:07:10.911352+00:00)

## What it is
SNIPER is an efficient multi-scale training algorithm for object detection and instance segmentation that processes only context regions (chips) around ground-truth objects instead of full image pyramids. AutoFocus is its companion inference algorithm that uses coarse-to-fine FocusPixels to only refine regions likely to contain small objects.

## Use cases
- train object detectors efficiently on multi-scale image pyramids
- speed up multi-scale inference for small object detection
- train instance segmentation models with large batch sizes on limited GPUs
- detect small objects in high-resolution images faster
- reproduce SNIPER and AutoFocus research results on COCO
- train detectors with batch normalization without cross-GPU synchronization

## When to choose
- you need efficient multi-scale training or inference for object detection or instance segmentation
- you want to reproduce published SNIPER/AutoFocus results on COCO
- you have GPU resources and want large-batch detector training on a single node
- small object detection speed and accuracy matter for your pipeline

## When to avoid
- you need a maintained production object detection framework - the code is research-grade with a nonstandard license and infrequent updates
- you want a simple pretrained detector API without deep learning research experience
- you work outside the CUDA/GPU Linux environment the code targets
- you need modern architectures or tooling beyond what this 2018-2021 research codebase supports

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, computer-vision, image-processing
- domain: computer-vision, deep-learning, machine-learning, image-processing
- platform: python
- tags: object-detection, instance-segmentation, multi-scale-training, multi-scale-inference, research-code, coco, neurips-2018, iccv-2019, linux, gpu, cuda

## Member repositories
- mahyarnajibi/SNIPER (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:10.911352+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 2026-08-30T02:16:35.028438+00:00, confidence not recorded.
  - readme: https://github.com/mahyarnajibi/SNIPER (fetched 2026-08-28T04:07:10.911352+00:00, sha 990d933b571c)
- Data as of 2026-08-30T08:39:29.467469+00:00.