python-vtonimat 0.12

Description:

pythonvtonimat 0.12

vton-imaterialist
A python package for Binary Segmentation DataSet ( vton_plus and imaterialist [topwear])
📚 PyPi Project Documentation 📚
Download dataset
Note - This step can be performed after installing package as well.
Download dataset from following drive and unzip it.
gdrive
Install package
Installation with pypi
pip3 install python-vtonimat

Installation from source
git clone https://github.com/aarti-b/vton-imaterialist
python3 setup.py install

Set path to use package outside directory
export PYTHONPATH="$PYTHONPATH:/path_to_github-clone-package/package/package/"


Usage Guide
There are two datasets this package focuses on

vton
imaterialist

vton dataset
default option for dataset is vton. Follow the following commands to load data. assign path value to the folder where data is downloaded and unzipped.
Load whole data
from vtonimat import SegData
images, labels = SegData(path='path_to_datafiles').load_training()

Load batchwise dataset
Load by batches. Following command returns list of batches. Batch size is input parameter in method load_training_in_batches.
Load whole data
from vton import SegData
images, labels = SegData().load_training_in_batches(1000)

imaterialist'19 topwear dataset
from vtonimat import SegData
images, labels = SegData(path='path_to_datafiles', dataset='imat19').load_training()

Load batchwise dataset
Load by batches. Following command returns list of batches. Batch size is input parameter in method load_training_in_batches.
from vton import SegData
images, labels = SegData().load_training_in_batches(1000)

There is a python file convert.py to convert dataset to ubyte format the dataset you downloaded from google drive link. This file converts 3D images and 2D labels images to ubyte format.
Usage to convert data
python3 convert.py train 0 #0 is ratio, which means whole data is converted to train. you can add proportions.
python3 convert.py test 0

This package is still in progress. If you find any issue please feel free to contact or create a new issue. You are welcome to contribute in this project.