CryoPARES

User Guide

  • Training Guide
    • Overview
    • Table of Contents
    • Quick Start
    • Training Parameters
      • Essential Parameters
      • Data Configuration Parameters
      • Optimizer Configuration
      • Model Architecture Parameters
      • Data Augmentation Parameters
    • Monitoring Training with TensorBoard
      • Launching TensorBoard
      • Key Metrics to Monitor
        • 1. Training Loss (loss)
        • 2. Validation Loss (val_loss)
        • 3. Angular Error (geo_degs, val_geo_degs)
        • 4. Median Angular Error (val_median_geo_degs)
        • 5. Learning Rate (lr)
        • 6. Visualization: Rotation Matrices
      • Example TensorBoard Monitoring Session
    • Overfitting and Underfitting
      • What is Overfitting?
      • What is Underfitting?
      • The Sweet Spot
    • Data Preprocessing
      • Image Size and Sampling Rate
      • Masking
      • CTF Correction
    • Advanced Training Options
      • Continue Training
      • Fine-tuning
      • Half-Set Training
      • Simulated Pre-training
      • Model Compilation
      • Debugging: Overfitting on Small Batches
    • Troubleshooting Training Issues
      • Training is very slow
      • Out of memory errors
      • “Too many open files” error
      • Loss is NaN
      • Model not improving
      • Validation loss jumps around
    • See Also
  • Configuration Guide
    • Table of Contents
    • Configuration System Overview
      • Configuration Sources (in order of precedence)
      • Using Config Overrides
      • Using YAML Config Files
    • Viewing Available Options
    • Configuration Hierarchy
    • Reading Configuration Files
      • Configuration File Locations
      • How to Read Configuration Files
      • Example: Finding Training Parameters
      • Example: Finding Nested Parameters
    • Quick Parameter Reference
    • Saving and Loading Configurations
      • Exporting Current Configuration
      • Reusing a Configuration
      • Overriding Loaded Config
    • See Also
  • Command-Line Interface
    • Table of Contents
    • cryopares_train
      • Usage
      • Parameters
      • Configuration Overrides
      • View All Config Options
      • Examples
      • Important Notes
      • See Also
    • cryopares_infer
      • Usage
      • Parameters
      • Configuration Overrides
      • Examples
      • Output Files
      • See Also
    • cryopares_projmatching
      • Usage
      • Parameters
      • Example
    • cryopares_reconstruct
      • Usage
      • Parameters
      • Example
    • compactify_checkpoint
      • Usage
      • Required Arguments
      • Optional Arguments
      • Examples
      • What’s Included
      • Size Reduction
      • Using Compactified Checkpoints
    • Utility Scripts
      • GMM Histogram Analysis
      • FSC Computation
      • Pose Comparison
      • Learning Curve Visualization
      • STAR File Histograms
    • See Also
  • Troubleshooting Guide
    • Table of Contents
    • Installation Issues
      • pip install fails with dependency conflicts
      • ImportError: No module named ‘cryoPARES’
      • CUDA version mismatch
    • Configuration Issues
      • Type mismatch errors when using –config
    • File System Issues
      • “Too many open files” error
      • Permission denied when writing outputs
      • Particle files not found
    • Memory Issues
      • Out of memory (OOM) during training
      • Out of memory during inference
      • RAM exhausted when loading data
    • Training Issues
      • Loss becomes NaN
      • Training doesn’t improve
      • Validation loss higher than training loss
      • Training crashes with “Killed”
      • Checkpoints not saving
    • Inference Issues
      • Predicted poses are random
      • Reconstruction is blurry
      • “No particles passed confidence threshold”
      • Inference slower than expected
    • Data Issues
      • CTF parameters missing
      • Orientation columns missing
      • Particles are poorly centered
      • Different sampling rate in data vs. reference
    • Performance Issues
      • Training is very slow
      • Inference is very slow
      • Data loading is bottleneck
    • CUDA/GPU Issues
      • “CUDA out of memory” but GPU seems empty
      • Multiple GPUs not being used
      • GPU slower than expected
    • Output Quality Issues
      • Reconstruction has artifacts
      • FSC is lower than expected
      • Reconstructed map has wrong hand
    • Getting More Help
      • Enable Debug Mode
      • Check Logs
      • Report Issues
    • See Also

API Reference

  • API Reference
    • Core Modules
      • Training API
        • Main Training Module
        • Training Execution
      • Inference API
        • Single Inferencer
        • Distributed Inference
        • Daemon Mode
      • Projection Matching API
        • Projection Matcher
        • Command-Line Interface
      • Reconstruction API
        • Reconstructor
        • Command-Line Interface
      • Data Management API
        • Data Manager
        • Particles Dataset
        • Augmentations
        • CTF Correction
      • Models API
        • PyTorch Lightning Model
        • Image2Sphere Network
        • Image Encoders
        • Directional Normalizer
      • Utilities API
        • Checkpoint Management
        • Path Utilities
        • PyTorch Utilities
        • Reconstruction Utilities
        • Geometry Utilities

Additional Resources

  • GitHub Repository
  • Paper
CryoPARES
  • Search

© Copyright 2025, Ruben Sanchez-Garcia.

Built with Sphinx using a theme provided by Read the Docs.