- Description:
This dataset is just like the CIFAR-10, except it has 100 classes containing 600 images each. There are 500 training images and 100 testing images per class. The 100 classes in the CIFAR-100 are grouped into 20 superclasses. Each image comes with a "fine" label (the class to which it belongs) and a "coarse" label (the superclass to which it belongs).
Additional Documentation: Explore on Papers With Code
Source code:
tfds.image_classification.Cifar100
Versions:
3.0.2
(default): No release notes.
Download size:
160.71 MiB
Dataset size:
132.03 MiB
Auto-cached (documentation): Yes
Splits:
Split | Examples |
---|---|
'test' |
10,000 |
'train' |
50,000 |
- Feature structure:
FeaturesDict({
'coarse_label': ClassLabel(shape=(), dtype=int64, num_classes=20),
'id': Text(shape=(), dtype=string),
'image': Image(shape=(32, 32, 3), dtype=uint8),
'label': ClassLabel(shape=(), dtype=int64, num_classes=100),
})
- Feature documentation:
Feature | Class | Shape | Dtype | Description |
---|---|---|---|---|
FeaturesDict | ||||
coarse_label | ClassLabel | int64 | ||
id | Text | string | ||
image | Image | (32, 32, 3) | uint8 | |
label | ClassLabel | int64 |
Supervised keys (See
as_supervised
doc):('image', 'label')
Figure (tfds.show_examples):
- Examples (tfds.as_dataframe):
- Citation:
@TECHREPORT{Krizhevsky09learningmultiple,
author = {Alex Krizhevsky},
title = {Learning multiple layers of features from tiny images},
institution = {},
year = {2009}
}