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


  • Description:

Caltech-UCSD Birds 200 (CUB-200) is an image dataset with photos of 200 bird species (mostly North American). The total number of categories of birds is 200 and there are 6033 images in the 2010 dataset and 11,788 images in the 2011 dataset. Annotations include bounding boxes, segmentation labels.

Split Examples
'test' 5,794
'train' 5,994
  • Feature structure:
    'bbox': BBoxFeature(shape=(4,), dtype=tf.float32),
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'image/filename': Text(shape=(), dtype=tf.string),
    'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=200),
    'label_name': Text(shape=(), dtype=tf.string),
    'segmentation_mask': Image(shape=(None, None, 1), dtype=tf.uint8),
  • Feature documentation:
Feature Class Shape Dtype Description
bbox BBoxFeature (4,) tf.float32
image Image (None, None, 3) tf.uint8
image/filename Text tf.string
label ClassLabel tf.int64
label_name Text tf.string
segmentation_mask Image (None, None, 1) tf.uint8
  • Citation:
Author = {P. Welinder and S. Branson and T. Mita and C. Wah and F. Schroff and S. Belongie and P. Perona},
Institution = {California Institute of Technology},
Number = {CNS-TR-2010-001},
Title = { {Caltech-UCSD Birds 200} },
Year = {2010}