pytorch-custom-utils's Introduction

pytorch-custom-utils

This is a lightweight repository to help PyTorch users.

Usage

📋 Dependencies

torch 1.4.0
torchvision 0.5.0
python 3.6
matplotlib 2.2.2
numpy 1.14.3
seaborn 0.9.0
sklearn
plotly

🔨 Installation

pip install torchhk or
git clone https://github.com/Harry24k/pytorch-custom-utils

from torchhk import *

🚀 Demos

RecordManager (code, markdown): RecordManager will help you to keep tracking training records.
Datasets (code, markdown): Dataset will help you to use torch datasets including split and label-filtering.

Supported datasets

# CIFAR10
datasets = Datasets("CIFAR10", root='./data')
    
# CIFAR100
datasets = Datasets("CIFAR100", root='./data')
    
# STL10
datasets = Datasets("STL10", root='./data')
    
# MNIST
datasets = Datasets("MNIST", root='./data')
    
# FashionMNIST
datasets = Datasets("FashionMNIST", root='./data')
    
# SVHN
datasets = Datasets("SVHN", root='./data')
    
# MNISTM
datasets = Datasets("MNISTM", root='./data')
    
# ImageNet
datasets = Datasets("ImageNet", root='./data')
    
# USPS
datasets = Datasets("USPS", root='./data')
    
# TinyImageNet
datasets = Datasets("TinyImageNet", root='./data')
    
# CIFAR with Unsupervised
datasets = Datasets("CIFAR10U", root='./data')
datasets = Datasets("CIFAR100U", root='./data')
    
# Corrupted CIFAR (Only test data will be corrupted)
# CORRUPTIONS = [
#    'gaussian_noise', 'shot_noise', 'impulse_noise', 'defocus_blur',
#    'glass_blur', 'motion_blur', 'zoom_blur', 'snow', 'frost', 'fog',
#    'brightness', 'contrast', 'elastic_transform', 'pixelate',
#    'jpeg_compression'
#]
datasets = Datasets("CIFAR10", root='./data',corruption='gaussian_noise')

Vis (code, markdown): Vis will help you to visualize torch tensors.
Transform (code): Transform will help you to change specific layers.