facebookresearch/detectron

★ 26,383PythonAudience · researcherComplexity · 4/5DormantLicenseSetup · hard

Mindmap

mindmap
  root((Detectron))
    What it does
      Object detection
      Instance segmentation
      Bounding boxes
      Pixel-level outlines
    Key algorithms
      Mask R-CNN
      Earlier detectors
    Use cases
      Photo background blur
      Security monitoring
      Medical imaging
      Autonomous vehicles
    Tech stack
      Python
      Deep learning
      Computer vision
    Status
      Deprecated
      Historical reference
      Detectron2 available

mindmap root((Detectron)) What it does Object detection Instance segmentation Bounding boxes Pixel-level outlines Key algorithms Mask R-CNN Earlier detectors Use cases Photo background blur Security monitoring Medical imaging Autonomous vehicles Tech stack Python Deep learning Computer vision Status Deprecated Historical reference Detectron2 available

Things people build with this

USE CASE 1

Study the original Mask R-CNN implementation and how instance segmentation algorithms work.

USE CASE 2

Build photo editing tools that automatically detect and blur backgrounds or isolate subjects.

USE CASE 3

Develop security camera systems that identify and track specific people or vehicles in footage.

USE CASE 4

Create medical imaging analysis tools that automatically detect and outline anomalies in scans.

Tech stack

PythonPyTorchCUDANumPy

Getting it running

Difficulty · hard Time to first run · 1h+

Requires CUDA/GPU setup and PyTorch compilation; multiple dependencies and model weights to download.

BSD 3-Clause license allows free use for research and commercial purposes with attribution.

In plain English

Detectron is Facebook AI Research's original platform for computer vision research, specifically for object detection, which is the AI task of identifying and locating objects within images and video. Think of it as the research workbench where Facebook's scientists developed and tested algorithms that could look at a photo and draw boxes around every person, car, dog, or other object in it. The most famous algorithm that came out of this platform is Mask R-CNN, which won a top prize at a major computer vision conference in 2017. Unlike earlier systems that only drew bounding boxes around objects, Mask R-CNN could also draw precise outlines around each object, distinguishing the exact pixels belonging to a person versus the background behind them. This technique is called instance segmentation and is now foundational to many real-world applications. For a non-technical founder, Detectron represents the origin point of a family of technologies now used in production systems everywhere: photo apps that blur backgrounds, security cameras that detect specific people or vehicles, medical imaging that identifies anomalies, and autonomous vehicles that identify pedestrians. Important note: this original Detectron is deprecated, meaning it's no longer actively maintained. The team rewrote it from scratch as Detectron2, which is the current version to use if you need this technology. This original repository remains publicly available as a historical reference and for anyone studying the original implementations of these landmark algorithms.

Copy-paste prompts

Prompt 1

How do I set up the original Detectron to run Mask R-CNN inference on my own images?

Prompt 2

Explain how Mask R-CNN differs from earlier object detection methods like Faster R-CNN in the Detectron codebase.

Prompt 3

I want to fine-tune a Detectron model on my custom dataset of objects, what's the training pipeline?

Prompt 4

Show me how to extract the instance segmentation masks from Detectron's output for post-processing.

Prompt 5

What are the key differences between the original Detectron and Detectron2, and should I migrate?

Open on GitHub → Explain another repo

Generated 2026-05-18 · Model: sonnet-4-6 · Verify against the repo before relying on details.