Octopi API Documentation¶

Octopi is a comprehensive 3D particle picking framework designed for cryo-electron tomography data analysis. This documentation covers the complete workflow from training to inference and evaluation.

Overview¶

Octopi provides a streamlined pipeline for:

• Training: Deep learning models for particle segmentation
• Inference: Automated particle detection and localization
• Evaluation: Performance assessment against ground truth annotations

Quick Start¶

For a minimal introduction to all core functions with essential parameters, see the Quick Start Guide. The sections below describe each component in greater detail.

Core Components¶

Configuration¶

All octopi workflows start with a Copick configuration file that defines:

• Data locations and formats
• Pickable object definitions with corresponding segmentation label values
• Tomogram metadata and processing parameters

The configuration file maps each pickable object to a specific integer value used in segmentation masks, enabling multi-class particle detection and classification.

Workflow Pages¶

Training¶

Learn how to:

• Create training targets from existing annotations
• Configure and train deep learning models
• Set up cross-validation splits
• Choose appropriate loss functions and model architectures

Inference¶

Discover how to:

• Run segmentation on new tomograms
• Perform particle localization from segmentation masks
• Configure test-time augmentation
• Evaluate results against ground truth