Pixel Reshaper
This package aims to supplement the image classification ML pipeline, particularly focused on PyTorch, by streamlining image generation from tabular data. It contains a simple combination of functions. First, to unpack an entire dataset from tabular to .png form, handling all train/test splits and file structures. Second, to provide a hand-ball function to prepare and image to be loaded for real time predicition, then move the image to archive after use.
Table of contents
Quick start
The package can be installed using pip
pip install pixel_reshaper
Usage
Import the package.
import pixel_reshaper as pxR
Next, some parameters relating to the naming of classes, and location for data to be unpacked to. These are used by the structure generation steps
#Classes in the data for image and directory naming
classNames = ['dog', 'cat', 'mouse']
#Location for the data to be unpacked to, and directory to create
loc = './'
dirName = 'unpacked_images'
The package pre-splits the data, so just supply a split proportion. Additionally, image dimensions and channels need to be specified (Auto dimensionality detection coming soon)
splitPercent = 0.25
dimImage = (27, 27, 3)
Lastly, specify the path the the dataset to be unpacked
fileName = './imageDataset.csv'
Pass all arguments to the function and the dataset will be unpacked to the desired location!
pxR.unpack_images(classNames, loc, dirName, splitPercent, dimImage, fileName)
Copyright and license
Code released under the MIT Licence.
