qlty

qlty is a Python library designed to handle large 2D or 3D tensors efficiently by splitting them into smaller, manageable chunks. This library is particularly useful for processing large datasets that do not fit into memory, enabling chunked processing for machine learning workflows.

Features

• Efficient tensor splitting and stitching

• Intelligent border handling to minimize artifacts

• Support for both in-memory and disk-cached processing

• 2D and 3D tensor support

• Sparse data handling utilities

• Numba acceleration for 2D stitching

Quick Start

Installation:

pip install qlty torch zarr numpy einops dask numba

Basic Usage:

import torch
from qlty import NCYXQuilt

# Create a quilt object
quilt = NCYXQuilt(
    Y=128, X=128,
    window=(32, 32),
    step=(16, 16),
    border=(5, 5),
    border_weight=0.1
)

# Split data into patches
data = torch.randn(10, 3, 128, 128)
patches = quilt.unstitch(data)

# Process patches (e.g., with a neural network)
processed = your_model(patches)

# Stitch back together
reconstructed, weights = quilt.stitch(processed)

Documentation

Full documentation is available at https://qlty.readthedocs.io

• Installation Guide

• Usage Guide

• Examples

• API Reference

• Troubleshooting

Modules

In-Memory Classes

• NCYXQuilt: 2D tensor splitting and stitching (shape: N, C, Y, X)

• NCZYXQuilt: 3D tensor splitting and stitching (shape: N, C, Z, Y, X)

Disk-Cached Classes

• LargeNCYXQuilt: 2D with on-disk caching using Zarr

• LargeNCZYXQuilt: 3D with on-disk caching using Zarr

License

• Free software: BSD license

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.