Join TensorFlow at Google I/O, May 11-12 Register now

aflw2k3d

  • Description:

AFLW2000-3D is a dataset of 2000 images that have been annotated with image-level 68-point 3D facial landmarks. This dataset is typically used for evaluation of 3D facial landmark detection models. The head poses are very diverse and often hard to be detected by a cnn-based face detector. The 2D landmarks are skipped in this dataset, since some of the data are not consistent to 21 points, as the original paper mentioned.

Split Examples
'train' 2,000
  • Feature structure:
FeaturesDict({
    'image': Image(shape=(450, 450, 3), dtype=tf.uint8),
    'landmarks_68_3d_xy_normalized': Tensor(shape=(68, 2), dtype=tf.float32),
    'landmarks_68_3d_z': Tensor(shape=(68, 1), dtype=tf.float32),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
image Image (450, 450, 3) tf.uint8
landmarks_68_3d_xy_normalized Tensor (68, 2) tf.float32
landmarks_68_3d_z Tensor (68, 1) tf.float32

Visualization

  • Citation:
@article{DBLP:journals/corr/ZhuLLSL15,
  author    = {Xiangyu Zhu and
               Zhen Lei and
               Xiaoming Liu and
               Hailin Shi and
               Stan Z. Li},
  title     = {Face Alignment Across Large Poses: {A} 3D Solution},
  journal   = {CoRR},
  volume    = {abs/1511.07212},
  year      = {2015},
  url       = {http://arxiv.org/abs/1511.07212},
  archivePrefix = {arXiv},
  eprint    = {1511.07212},
  timestamp = {Mon, 13 Aug 2018 16:48:23 +0200},
  biburl    = {https://dblp.org/rec/bib/journals/corr/ZhuLLSL15},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}