ucf101

  • Description:

A 101-label video classification dataset.

@article{DBLP:journals/corr/abs-1212-0402,
  author    = {Khurram Soomro and
               Amir Roshan Zamir and
               Mubarak Shah},
  title     = { {UCF101:} {A} Dataset of 101 Human Actions Classes From Videos in
               The Wild},
  journal   = {CoRR},
  volume    = {abs/1212.0402},
  year      = {2012},
  url       = {http://arxiv.org/abs/1212.0402},
  archivePrefix = {arXiv},
  eprint    = {1212.0402},
  timestamp = {Mon, 13 Aug 2018 16:47:45 +0200},
  biburl    = {https://dblp.org/rec/bib/journals/corr/abs-1212-0402},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

ucf101/ucf101_1_256 (default config)

  • Config description: 256x256 UCF with the first action recognition split.

  • Dataset size: 7.40 GiB

  • Splits:

Split Examples
'test' 3,783
'train' 9,537
  • Feature structure:
FeaturesDict({
    'label': ClassLabel(shape=(), dtype=int64, num_classes=101),
    'video': Video(Image(shape=(256, 256, 3), dtype=uint8)),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
label ClassLabel int64
video Video(Image) (None, 256, 256, 3) uint8

ucf101/ucf101_1

  • Config description: UCF with the action recognition split #1.

  • Dataset size: 8.48 GiB

  • Splits:

Split Examples
'test' 3,783
'train' 9,537
  • Feature structure:
FeaturesDict({
    'label': ClassLabel(shape=(), dtype=int64, num_classes=101),
    'video': Video(Image(shape=(None, None, 3), dtype=uint8)),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
label ClassLabel int64
video Video(Image) (None, None, None, 3) uint8

ucf101/ucf101_2

  • Config description: UCF with the action recognition split #2.

  • Dataset size: 8.48 GiB

  • Splits:

Split Examples
'test' 3,734
'train' 9,586
  • Feature structure:
FeaturesDict({
    'label': ClassLabel(shape=(), dtype=int64, num_classes=101),
    'video': Video(Image(shape=(None, None, 3), dtype=uint8)),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
label ClassLabel int64
video Video(Image) (None, None, None, 3) uint8

ucf101/ucf101_3

  • Config description: UCF with the action recognition split #3.

  • Dataset size: 8.48 GiB

  • Splits:

Split Examples
'test' 3,696
'train' 9,624
  • Feature structure:
FeaturesDict({
    'label': ClassLabel(shape=(), dtype=int64, num_classes=101),
    'video': Video(Image(shape=(None, None, 3), dtype=uint8)),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
label ClassLabel int64
video Video(Image) (None, None, None, 3) uint8