Google I/O is a wrap! Catch up on TensorFlow sessions View sessions


  • Description:

ImageNet-Sketch consists of 50,889 black and white sketch images, 50 for each of the 1000 ImageNet classes. These images were originally collected from Google Image Search for "sketch of __". 100 images were collected and then manually filtered. For classes with fewer than 50 good images, additional images were constructed by flip or rotation.

Split Examples
'test' 50,889
  • Feature structure:
    'file_name': Text(shape=(), dtype=tf.string),
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=1000),
  • Feature documentation:
Feature Class Shape Dtype Description
file_name Text tf.string
image Image (None, None, 3) tf.uint8
label ClassLabel tf.int64


  • Citation:
        title={Learning Robust Global Representations by Penalizing Local Predictive Power},
        author={Wang, Haohan and Ge, Songwei and Lipton, Zachary and Xing, Eric P},
        booktitle={Advances in Neural Information Processing Systems},