Tutorials

• Installation
  • Requirements
  • Build Detectron2 from Source
  • Install Pre-Built Detectron2 (Linux only)
  • Common Installation Issues
  • Installation inside specific environments:
• Getting Started with Detectron2
  • Inference Demo with Pre-trained Models
  • Training & Evaluation in Command Line
  • Use Detectron2 APIs in Your Code
• Use Builtin Datasets
  • Expected dataset structure for COCO instance/keypoint detection:
  • Expected dataset structure for PanopticFPN:
  • Expected dataset structure for LVIS instance segmentation:
  • Expected dataset structure for cityscapes:
  • Expected dataset structure for Pascal VOC:
  • Expected dataset structure for ADE20k Scene Parsing:
• Extend Detectron2’s Defaults
• Use Custom Datasets
  • Register a Dataset
  • “Metadata” for Datasets
  • Register a COCO Format Dataset
  • Update the Config for New Datasets
• Dataloader
  • How the Existing Dataloader Works
  • Write a Custom Dataloader
  • Use a Custom Dataloader
• Data Augmentation
  • Basic Usage
  • Write New Augmentations
  • Advanced Usage
• Use Models
  • Build Models from Yacs Config
• Write Models
  • Register New Components
  • Construct Models with Explicit Arguments
• Training
  • Custom Training Loop
  • Trainer Abstraction
  • Logging of Metrics
• Evaluation
  • Use evaluators
  • Evaluators for custom dataset
• Yacs Configs
  • Basic Usage
  • Configs in Projects
  • Best Practice with Configs
• Lazy Configs
  • Python Syntax
  • Recursive Instantiation
  • Using Model Zoo LazyConfigs
  • Summary
• Deployment
  • Deployment with Tracing or Scripting
  • Deployment with Caffe2-tracing
  • Conversion to TensorFlow