Package structure

quantEM is organized as a core infrastructure package plus one subpackage per experimental modality. Core holds everything shared: data structures, file I/O, visualization, configuration, fitting, and machine learning building blocks. Each modality subpackage then holds the analysis specific to one kind of measurement.

quantem
├── core
│   ├── datastructures
│   ├── io
│   ├── visualization
│   ├── config
│   ├── fitting
│   ├── ml
│   └── utils
├── imaging
├── diffraction
├── diffractive_imaging
├── tomography
└── spectroscopy

Inside core: datastructures holds the Dataset classes and Vector; io holds the file readers and Zarr serialization; visualization holds show_2d, linescan, colormaps, and scale bars; config holds the global settings for device, precision, and plotting defaults; fitting holds least-squares fitting, backgrounds, and diffraction models; ml holds CNNs, autoencoders, implicit neural representations, loss functions, and distributed training; utils holds array helpers, filters, validators, and random number generation.

The modality subpackages cover, in order: real-space images, diffraction-space measurements, phase retrieval, tilt series, and spectra.

What each modality covers¶

quantem.imaging works on real-space S/TEM images and image series.

• drift: measure and correct scan drift and distortion across an image series

• lattice: fit atomic lattices and map local lattice distortions

quantem.diffraction works on diffraction patterns, mainly nanobeam and 4D-STEM.

• disk_detection, bragg_vectors: locate Bragg disks at each probe position and store them as ragged Vector data

• strain, strain_autocorrelation: convert fitted lattice vectors into strain and rotation maps

• model_fitting: fit a forward model directly to the diffraction patterns, as an alternative to disk detection

quantem.diffractive_imaging recovers phase from 4D-STEM data.

• ptychography, ptychography_lite: iterative gradient-based reconstruction, from quick previews to multi-GPU runs

• direct_ptychography: non-iterative reconstruction

• interchangeable model components: object_models, probe_models, detector_models, dataset_models, origin_models

• constraints, optimize_hyperparameters: regularization and Optuna-driven hyperparameter search

quantem.tomography reconstructs 3D volumes from tilt series.

• tomography, tomography_lite: reconstruction, including implicit neural representations

• preprocess: alignment and preprocessing of the tilt series

• radon: forward and adjoint projection operators

quantem.spectroscopy analyzes spectroscopic signals, principally energy-dispersive X-ray spectroscopy (XEDS) and electron energy loss spectroscopy (EELS).

Import conventions¶

The top-level package re-exports the pieces most workflows need:

import quantem

quantem.io
quantem.datastructures
quantem.visualization

These give the file readers and load/save, the Dataset classes, and the plotting helpers.

Modality subpackages are imported explicitly, for example:

from quantem.diffractive_imaging import ptychography
from quantem.core import config

Datasets carry calibrations¶

All analysis operates on Dataset objects: thin wrappers around an array (NumPy or torch tensor) plus calibration metadata such as sampling, units, origin, name, and signal_units. Dimension-specific subclasses (Dataset2d, Dataset3d, Dataset4d, Dataset4dstem) add modality-specific functionality, and Vector handles ragged point-like data such as detected Bragg peaks. See Datasets & I/O.

Relationship to the tutorials¶

The quantem-tutorials repository contains runnable Jupyter notebooks organized by the same modality names as the subpackages. The User Guide pages in this documentation are condensed versions of those notebooks. For a complete, executable walkthrough with data, use the tutorials.