Data zoo
This repository provides unified access to multiple datasets.
Usage
First of all, you have to import data_provider from datazoo package:
from datazoo import data_provider
Then, you can select dataset from the list and get iterable:
# Dataset object
fashionmnist = data_provider(
dataset='fashionmnist', data_dir='data/fashionmnist/', split='test',
download=True, columns=['index', 'image', 'class']
)
print('Dataset length:', len(fashionmnist))
# Iterate over samples
for i in fashionmnist:
print(i)
Classification
Single-label datasets
| Dataset | Name in data provider | Number of classes | Number of samples | Source | Auto downloading |
|---|---|---|---|---|---|
| MNIST | mnist |
10 | 60 000 / 10 000 | torchvision | Yes |
| Fashion MNIST | fashionmnist |
10 | 60 000 / 10 000 | torchvision | Yes |
| CIFAR-10 | cifar10 |
10 | 50 000 / 10 000 | torchvision | Yes |
| CIFAR-100 | cifar100 |
100 | 50 000 / 10 000 | torchvision | Yes |
| Indoor Scene Recognition | indoor_scene_recon |
67 | 15620 | -- | Yes |
| The Street View House Numbers (SVHN) | svhn_cropped |
10 | 73257 digits for training, 26032 digits for testing, and 531131 additional | -- | Yes |
| Linnaeus5 | linnaeus5 |
5 classes: berry, bird, dog, flower, other (negative set) | 1200 training images, 400 test images per class | -- | Yes |
| COIL-100 | coil100 |
100 (100 objects) | 7200 images | -- | Yes |
License
This software is covered by MIT License.
