# JosephKJ/OWOD

(CVPR 2021 Oral) Open World Object Detection

Repository: https://github.com/JosephKJ/OWOD
Canonical: https://ross.abutalabs.com/products/owod
Homepage: https://josephkj.in
Language: Python
License: Apache-2.0
License Family: permissive
Topics: open-world, object-detection, contrastive-learning, energy-based-model, detectron2, incremental-learning, continual-learning, cvpr2021, cvpr
Last push: 2022-12-19T08:59:57+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": 2165, "days_push": 1353, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1074, forks 152 (observed 2026-08-28T04:03:29.037721+00:00)

## What it is
OWOD is the official PyTorch implementation of ORE (Open World Object Detector), a CVPR 2021 Oral paper introducing the Open World Object Detection problem. Built on Detectron2, it detects unknown object classes without explicit supervision and incrementally learns new classes via contrastive clustering and energy-based unknown identification.

## Use cases
- detect objects from classes the model was never trained on
- incrementally add new object classes to a detector without forgetting old ones
- reproduce CVPR 2021 open world object detection experiments
- research continual learning for object detection
- identify unknown instances in images as 'unknown' instead of misclassifying them
- benchmark open-set and open-world detection on COCO-style datasets

## When to choose
- you need a research baseline for open-world or open-set object detection
- you want to study incremental/continual object detection with unknown identification
- you already use Detectron2 and want to extend it with energy-based unknown detection
- you need the reference implementation of the ORE paper

## When to avoid
- you need a production-ready object detector with a maintained ecosystem
- you want plug-and-play detection of a fixed set of classes (use standard Detectron2 or YOLO)
- you need Windows support or easy installation without Detectron2 build steps
- you expect frequent updates or active maintenance

## Facets
- artifact type: library
- maturity: maintenance
- function: computer-vision, machine-learning, deep-learning
- domain: computer-vision, machine-learning, deep-learning
- platform: python
- tags: object-detection, open-world, continual-learning, incremental-learning, contrastive-learning, energy-based-models, detectron2, research-code, cvpr2021, linux, gpu

## Member repositories
- JosephKJ/OWOD (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:29.037721+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 2026-08-30T06:53:29.869010+00:00, confidence not recorded.
  - readme: https://github.com/JosephKJ/OWOD (fetched 2026-08-28T04:03:29.037721+00:00, sha b781eebb7b96)
  - homepage: https://josephkj.in (fetched 2026-08-29T12:55:29.759367+00:00, sha 3ffa42e9bd5a)
- Data as of 2026-08-30T08:39:29.467469+00:00.