visual_domain_decathlon

This contains the 10 datasets used in the Visual Domain Decathlon, part of the PASCAL in Detail Workshop Challenge (CVPR 2017). The goal of this challenge is to solve simultaneously ten image classification problems representative of very different visual domains.

Some of the datasets included here are also available as separate datasets in TFDS. However, notice that images were preprocessed for the Visual Domain Decathlon (resized isotropically to have a shorter size of 72 pixels) and might have different train/validation/test splits. Here we use the official splits for the competition.

visual_domain_decathlon is configured with tfds.image.visual_domain_decathlon.VisualDomainDecathlonConfig and has the following configurations predefined (defaults to the first one):

  • aircraft (v1.1.0) (Size: 1.04 GiB): Data based on "Aircraft", with images resized isotropically to have a shorter size of 72 pixels.

  • cifar100 (v1.1.0) (Size: 1.04 GiB): Data based on "CIFAR-100", with images resized isotropically to have a shorter size of 72 pixels.

  • daimlerpedcls (v1.1.0) (Size: 1.04 GiB): Data based on "Daimler Pedestrian Classification", with images resized isotropically to have a shorter size of 72 pixels.

  • dtd (v1.1.0) (Size: 1.04 GiB): Data based on "Describable Textures", with images resized isotropically to have a shorter size of 72 pixels.

  • gtsrb (v1.1.0) (Size: 1.04 GiB): Data based on "German Traffic Signs", with images resized isotropically to have a shorter size of 72 pixels.

  • imagenet12 (v1.1.0) (Size: 6.40 GiB): Data based on "Imagenet", with images resized isotropically to have a shorter size of 72 pixels.

  • omniglot (v1.1.0) (Size: 1.04 GiB): Data based on "Omniglot", with images resized isotropically to have a shorter size of 72 pixels.

  • svhn (v1.1.0) (Size: 1.04 GiB): Data based on "Street View House Numbers", with images resized isotropically to have a shorter size of 72 pixels.

  • ucf101 (v1.1.0) (Size: 1.04 GiB): Data based on "UCF101 Dynamic Images", with images resized isotropically to have a shorter size of 72 pixels.

  • vgg-flowers (v1.1.0) (Size: 1.04 GiB): Data based on "VGG-Flowers", with images resized isotropically to have a shorter size of 72 pixels.

visual_domain_decathlon/aircraft

Data based on "Aircraft", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 10,000
TRAIN 3,334
TEST 3,333
VALIDATION 3,333

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=100),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

visual_domain_decathlon/cifar100

Data based on "CIFAR-100", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 60,000
TRAIN 40,000
TEST 10,000
VALIDATION 10,000

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=100),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

visual_domain_decathlon/daimlerpedcls

Data based on "Daimler Pedestrian Classification", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 49,000
TRAIN 23,520
TEST 19,600
VALIDATION 5,880

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=2),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

visual_domain_decathlon/dtd

Data based on "Describable Textures", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 5,640
TEST 1,880
TRAIN 1,880
VALIDATION 1,880

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=47),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

visual_domain_decathlon/gtsrb

Data based on "German Traffic Signs", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 51,839
TRAIN 31,367
TEST 12,630
VALIDATION 7,842

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=43),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

visual_domain_decathlon/imagenet12

Data based on "Imagenet", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 1,329,405
TRAIN 1,232,167
VALIDATION 49,000
TEST 48,238

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=1000),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

visual_domain_decathlon/omniglot

Data based on "Omniglot", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 32,460
TRAIN 17,853
TEST 8,115
VALIDATION 6,492

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=1623),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

visual_domain_decathlon/svhn

Data based on "Street View House Numbers", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 99,289
TRAIN 47,217
VALIDATION 26,040
TEST 26,032

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=10),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

visual_domain_decathlon/ucf101

Data based on "UCF101 Dynamic Images", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 13,320
TRAIN 7,585
TEST 3,783
VALIDATION 1,952

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=101),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

visual_domain_decathlon/vgg-flowers

Data based on "VGG-Flowers", with images resized isotropically to have a shorter size of 72 pixels.

Versions:

  • 1.1.0 (default):

Statistics

Split Examples
ALL 8,189
TEST 6,149
TRAIN 1,020
VALIDATION 1,020

Features

FeaturesDict({
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=102),
    'name': Text(shape=(), dtype=tf.string),
})

Homepage

Supervised keys (for as_supervised=True)

(u'image', u'label')

Citation

@ONLINE{hakanbilensylvestrerebuffitomasjakab2017,
    author = "Hakan Bilen, Sylvestre Rebuffi, Tomas Jakab",
    title  = "Visual Domain Decathlon",
    year   = "2017",
    url    = "https://www.robots.ox.ac.uk/~vgg/decathlon/"
}