moving_mnist

  • Description:

Moving variant of MNIST database of handwritten digits. This is the data used by the authors for reporting model performance. See tfds.video.moving_mnist.image_as_moving_sequence for generating training/validation data from the MNIST dataset.

Split Examples
'test' 10,000
  • Feature structure:
FeaturesDict({
    'image_sequence': Video(Image(shape=(64, 64, 1), dtype=uint8)),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
image_sequence Video(Image) (20, 64, 64, 1) uint8
  • Citation:
@article{DBLP:journals/corr/SrivastavaMS15,
  author    = {Nitish Srivastava and
               Elman Mansimov and
               Ruslan Salakhutdinov},
  title     = {Unsupervised Learning of Video Representations using LSTMs},
  journal   = {CoRR},
  volume    = {abs/1502.04681},
  year      = {2015},
  url       = {http://arxiv.org/abs/1502.04681},
  archivePrefix = {arXiv},
  eprint    = {1502.04681},
  timestamp = {Mon, 13 Aug 2018 16:47:05 +0200},
  biburl    = {https://dblp.org/rec/bib/journals/corr/SrivastavaMS15},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}