glow 0.10.5

Toolset for model training and creation of pipelines

Glow Library

Collection of tools for easier prototyping with deep learning extensions (PyTorch framework)

Overview

...

Installation

For basic installation use:

pip install glow

Specific versions with additional requirements

pip install glow[nn]  # For cv/neural network extras
pip install glow[io]  # For I/O extras
pip install glow[all]  # For all

Glow is compatible with: Python 3.9+, PyTorch 1.9+. Tested on ArchLinux, Ubuntu 18.04/20.04, Windows 10.

Structure

• glow.* - Core parts, available out the box
• glow.cv.* - Tools for computer vision tasks
• glow.io.* - I/O wrappers to access data in convenient formats
• glow.transforms - Some custom-made augmentations for data
• glow.nn - Neural nets and building blocks for them
• glow.metrics - Metric to use while training your neural network

Core features

• glow.mapped - convenient tool to parallelize computations
• glow.memoize - use if you want to reduce number of calls for any function

IO features

glow.io.TiledImage - ndarray-like reader for multiscale images (svs, tiff, etc...)

CTypes-based replacement of torchslide (deprecated).

from glow.io import read_tiled

slide = read_tiled('test.svs')
shape: tuple[int, ...] = slide.shape
scales: tuple[int, ...] = slide.scales
image: np.ndarray = slide[:2048, :2048]  # Get numpy.ndarray

glow.io.Sound - playable sound wrapper

from datetime import timedelta

import numpy as np
from glow.io import Sound

array: np.ndarray
sound = Sound(array, rate=44100)  # Wrap np.ndarray
sound = Sound.load('test.flac')  # Load sound into memory from file

# Get properties
rate: int = sound.rate
duration: timedelta = sound.duration
dtype: np.dtype = sound.dtype

 # Plays sound through default device, supports Ctrl-C for interruption
sound.play()