# facebookresearch/detectron2 architecture

> Architecture diagram of the GitHub repository facebookresearch/detectron2, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-09). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/facebookresearch/detectron2
- Repository: https://github.com/facebookresearch/detectron2

## Overview

Detectron2 is a modular PyTorch computer-vision library for detection and segmentation, used through configurable training, inference, evaluation, and deployment workflows. The main path is configuration and dataset preparation into a trainer, model, and evaluation; pretrained model access and export are additional capabilities. The map emphasizes components and connections supported by the README and source index. Some model and data internals were not sampled, so uncertain links are omitted.

Main entry points: [`detectron2/engine/defaults.py`](https://github.com/facebookresearch/detectron2/blob/main/detectron2/engine/defaults.py) (Default trainer), [`projects`](https://github.com/facebookresearch/detectron2/tree/main/projects) (Research projects), [`detectron2/modeling/meta_arch`](https://github.com/facebookresearch/detectron2/tree/main/detectron2/modeling/meta_arch) (Model architecture), [`detectron2/modeling/roi_heads/roi_heads.py`](https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/roi_heads/roi_heads.py) (ROI heads)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_workflow["Training workflow"]
  node_config["Configuration<br/>[config.py]"]
  node_datasets["Dataset registry"]
  node_loader["Data loading<br/>[build.py]"]
  node_trainer["Default trainer<br/>[defaults.py]"]
  node_trainloop["Training loop<br/>[train_loop.py]"]
  node_hooks["Training hooks<br/>[hooks.py]"]
  node_events["Metrics and events<br/>[events.py]"]
  node_solver["Optimizer and schedule<br/>[build.py]"]
end

subgraph group_vision["Vision modeling"]
  node_model["Model architecture"]
  node_backbone["Feature backbones"]
  node_proposals["Proposal generation"]
  node_roi["ROI heads<br/>[roi_heads.py]"]
  node_maskhead["Mask prediction<br/>[mask_head.py]"]
  node_structures["Vision structures<br/>[boxes.py]"]
end

subgraph group_results["Evaluation and deployment"]
  node_evaluator["COCO evaluation<br/>[coco_evaluation.py]"]
  node_checkpoint["Model checkpoints"]
  node_zoo["Model zoo<br/>[model_zoo.py]"]
  node_export["Model export"]
end

subgraph group_extensions["Research extensions"]
  node_extensions["Research projects"]
  node_densepose["DensePose<br/>[inference.py]"]
  node_panoptic["Panoptic DeepLab<br/>[panoptic_seg.py]"]
  node_pointrend["PointRend<br/>[mask_head.py]"]
  node_tensormask["TensorMask<br/>[arch.py]"]
end

node_user(("Researcher or app"))

node_user -->|"sets up"| node_config
node_user -->|"starts training"| node_trainer
node_config -->|"configures"| node_trainer
node_trainer -->|"builds"| node_model
node_trainer -->|"builds"| node_loader
node_datasets -->|"provides datasets"| node_loader
node_trainer -->|"delegates training"| node_trainloop
node_trainer -->|"builds optimizer"| node_solver
node_hooks -->|"observes loop"| node_trainloop
node_hooks -->|"writes metrics"| node_events
node_model -.->|"uses features"| node_backbone
node_model -.->|"generates proposals"| node_proposals
node_model -.->|"dispatches regions"| node_roi
node_roi -->|"builds mask head"| node_maskhead
node_roi -->|"uses structures"| node_structures
node_trainer -->|"evaluates models"| node_evaluator
node_evaluator -->|"converts instances"| node_structures
node_trainer -->|"loads weights"| node_checkpoint
node_user -->|"selects pretrained models"| node_zoo
node_zoo -->|"loads model weights"| node_checkpoint
node_user -->|"exports model"| node_export
node_user -->|"uses extensions"| node_extensions
node_extensions -->|"includes"| node_densepose
node_extensions -->|"includes"| node_panoptic
node_extensions -->|"includes"| node_pointrend
node_extensions -->|"includes"| node_tensormask

click node_config "https://github.com/facebookresearch/detectron2/blob/main/detectron2/config/config.py"
click node_datasets "https://github.com/facebookresearch/detectron2/tree/main/detectron2/data/datasets"
click node_loader "https://github.com/facebookresearch/detectron2/blob/main/detectron2/data/build.py"
click node_trainer "https://github.com/facebookresearch/detectron2/blob/main/detectron2/engine/defaults.py"
click node_trainloop "https://github.com/facebookresearch/detectron2/blob/main/detectron2/engine/train_loop.py"
click node_hooks "https://github.com/facebookresearch/detectron2/blob/main/detectron2/engine/hooks.py"
click node_events "https://github.com/facebookresearch/detectron2/blob/main/detectron2/utils/events.py"
click node_solver "https://github.com/facebookresearch/detectron2/blob/main/detectron2/solver/build.py"
click node_model "https://github.com/facebookresearch/detectron2/tree/main/detectron2/modeling/meta_arch"
click node_backbone "https://github.com/facebookresearch/detectron2/tree/main/detectron2/modeling/backbone"
click node_proposals "https://github.com/facebookresearch/detectron2/tree/main/detectron2/modeling/proposal_generator"
click node_roi "https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/roi_heads/roi_heads.py"
click node_maskhead "https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/roi_heads/mask_head.py"
click node_structures "https://github.com/facebookresearch/detectron2/blob/main/detectron2/structures/boxes.py"
click node_evaluator "https://github.com/facebookresearch/detectron2/blob/main/detectron2/evaluation/coco_evaluation.py"
click node_checkpoint "https://github.com/facebookresearch/detectron2/blob/main/detectron2/checkpoint/detection_checkpoint.py"
click node_zoo "https://github.com/facebookresearch/detectron2/blob/main/detectron2/model_zoo/model_zoo.py"
click node_export "https://github.com/facebookresearch/detectron2/tree/main/detectron2/export"
click node_extensions "https://github.com/facebookresearch/detectron2/tree/main/projects"
click node_densepose "https://github.com/facebookresearch/detectron2/blob/main/projects/DensePose/densepose/modeling/inference.py"
click node_panoptic "https://github.com/facebookresearch/detectron2/blob/main/projects/Panoptic-DeepLab/panoptic_deeplab/panoptic_seg.py"
click node_pointrend "https://github.com/facebookresearch/detectron2/blob/main/projects/PointRend/point_rend/mask_head.py"
click node_tensormask "https://github.com/facebookresearch/detectron2/blob/main/projects/TensorMask/tensormask/arch.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_config,node_datasets,node_loader,node_trainer,node_trainloop,node_hooks,node_events,node_solver toneBlue
class node_model,node_backbone,node_proposals,node_roi,node_maskhead,node_structures toneAmber
class node_evaluator,node_checkpoint,node_zoo,node_export toneMint
class node_extensions,node_densepose,node_panoptic,node_pointrend,node_tensormask toneRose
class node_user toneIndigo
```

## Components

### Training workflow

- [Configuration](https://github.com/facebookresearch/detectron2/blob/main/detectron2/config/config.py): `detectron2/config/config.py`
- [Dataset registry](https://github.com/facebookresearch/detectron2/tree/main/detectron2/data/datasets): `detectron2/data/datasets`
- [Data loading](https://github.com/facebookresearch/detectron2/blob/main/detectron2/data/build.py): `detectron2/data/build.py`
- [Default trainer](https://github.com/facebookresearch/detectron2/blob/main/detectron2/engine/defaults.py): `detectron2/engine/defaults.py`
- [Training loop](https://github.com/facebookresearch/detectron2/blob/main/detectron2/engine/train_loop.py): `detectron2/engine/train_loop.py`
- [Training hooks](https://github.com/facebookresearch/detectron2/blob/main/detectron2/engine/hooks.py): `detectron2/engine/hooks.py`
- [Metrics and events](https://github.com/facebookresearch/detectron2/blob/main/detectron2/utils/events.py): `detectron2/utils/events.py`
- [Optimizer and schedule](https://github.com/facebookresearch/detectron2/blob/main/detectron2/solver/build.py): `detectron2/solver/build.py`

### Vision modeling

- [Model architecture](https://github.com/facebookresearch/detectron2/tree/main/detectron2/modeling/meta_arch): `detectron2/modeling/meta_arch`
- [Feature backbones](https://github.com/facebookresearch/detectron2/tree/main/detectron2/modeling/backbone): `detectron2/modeling/backbone`
- [Proposal generation](https://github.com/facebookresearch/detectron2/tree/main/detectron2/modeling/proposal_generator): `detectron2/modeling/proposal_generator`
- [ROI heads](https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/roi_heads/roi_heads.py): `detectron2/modeling/roi_heads/roi_heads.py`
- [Mask prediction](https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/roi_heads/mask_head.py): `detectron2/modeling/roi_heads/mask_head.py`
- [Vision structures](https://github.com/facebookresearch/detectron2/blob/main/detectron2/structures/boxes.py): `detectron2/structures/boxes.py`

### Evaluation and deployment

- [COCO evaluation](https://github.com/facebookresearch/detectron2/blob/main/detectron2/evaluation/coco_evaluation.py): `detectron2/evaluation/coco_evaluation.py`
- [Model checkpoints](https://github.com/facebookresearch/detectron2/blob/main/detectron2/checkpoint/detection_checkpoint.py): `detectron2/checkpoint/detection_checkpoint.py`
- [Model zoo](https://github.com/facebookresearch/detectron2/blob/main/detectron2/model_zoo/model_zoo.py): `detectron2/model_zoo/model_zoo.py`
- [Model export](https://github.com/facebookresearch/detectron2/tree/main/detectron2/export): `detectron2/export`

### Research extensions

- [Research projects](https://github.com/facebookresearch/detectron2/tree/main/projects): `projects`
- [DensePose](https://github.com/facebookresearch/detectron2/blob/main/projects/DensePose/densepose/modeling/inference.py): `projects/DensePose/densepose/modeling/inference.py`
- [Panoptic DeepLab](https://github.com/facebookresearch/detectron2/blob/main/projects/Panoptic-DeepLab/panoptic_deeplab/panoptic_seg.py): `projects/Panoptic-DeepLab/panoptic_deeplab/panoptic_seg.py`
- [PointRend](https://github.com/facebookresearch/detectron2/blob/main/projects/PointRend/point_rend/mask_head.py): `projects/PointRend/point_rend/mask_head.py`
- [TensorMask](https://github.com/facebookresearch/detectron2/blob/main/projects/TensorMask/tensormask/arch.py): `projects/TensorMask/tensormask/arch.py`

### Other components

- **Researcher or app**

## Connections

- Researcher or app → Configuration: sets up
- Researcher or app → Default trainer: starts training
- Configuration → Default trainer: configures (evidence: `detectron2/engine/defaults.py`)
- Default trainer → Model architecture: builds (evidence: `detectron2/engine/defaults.py`)
- Default trainer → Data loading: builds (evidence: `detectron2/engine/defaults.py`)
- Dataset registry → Data loading: provides datasets (evidence: `detectron2/data/__init__.py`)
- Default trainer → Training loop: delegates training (evidence: `detectron2/engine/defaults.py`)
- Default trainer → Optimizer and schedule: builds optimizer (evidence: `detectron2/engine/defaults.py`)
- Training hooks → Training loop: observes loop (evidence: `detectron2/engine/hooks.py`)
- Training hooks → Metrics and events: writes metrics (evidence: `detectron2/engine/hooks.py`)
- Model architecture → Feature backbones: uses features
- Model architecture → Proposal generation: generates proposals
- Model architecture → ROI heads: dispatches regions
- ROI heads → Mask prediction: builds mask head (evidence: `detectron2/modeling/roi_heads/roi_heads.py`)
- ROI heads → Vision structures: uses structures (evidence: `detectron2/modeling/roi_heads/roi_heads.py`)
- Default trainer → COCO evaluation: evaluates models (evidence: `detectron2/engine/defaults.py`)
- COCO evaluation → Vision structures: converts instances (evidence: `detectron2/evaluation/coco_evaluation.py`)
- Default trainer → Model checkpoints: loads weights (evidence: `detectron2/engine/defaults.py`)
- Researcher or app → Model zoo: selects pretrained models
- Model zoo → Model checkpoints: loads model weights (evidence: `detectron2/model_zoo/model_zoo.py`)
- Researcher or app → Model export: exports model
- Researcher or app → Research projects: uses extensions
- Research projects → DensePose: includes
- Research projects → Panoptic DeepLab: includes (evidence: `projects/Panoptic-DeepLab/panoptic_deeplab/panoptic_seg.py`)
- Research projects → PointRend: includes (evidence: `projects/PointRend/point_rend/mask_head.py`)
- Research projects → TensorMask: includes (evidence: `projects/TensorMask/tensormask/arch.py`)
