COCO is a large-scale object detection, segmentation, and captioning dataset. This version contains images, bounding boxes, labels, and captions from COCO 2014, split into the subsets defined by Karpathy and Li (2015). This effectively divides the original COCO 2014 validation data into new 5000-image validation and test sets, plus a "restval" set containing the remaining ~30k images. All splits have caption annotations.

Split Examples
'restval' 30,504
'test' 5,000
'train' 82,783
'val' 5,000
  • Feature structure:
    'captions': Sequence({
        'id': tf.int64,
        'text': tf.string,
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'image/filename': Text(shape=(), dtype=tf.string),
    'image/id': tf.int64,
    'objects': Sequence({
        'area': tf.int64,
        'bbox': BBoxFeature(shape=(4,), dtype=tf.float32),
        'id': tf.int64,
        'is_crowd': tf.bool,
        'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=80),
  • Feature documentation:
Feature Class Shape Dtype Description
captions Sequence
captions/id Tensor tf.int64
captions/text Tensor tf.string
image Image (None, None, 3) tf.uint8
image/filename Text tf.string
image/id Tensor tf.int64
objects Sequence
objects/area Tensor tf.int64
objects/bbox BBoxFeature (4,) tf.float32
objects/id Tensor tf.int64
objects/is_crowd Tensor tf.bool
objects/label ClassLabel tf.int64
  author    = {Tsung{-}Yi Lin and
               Michael Maire and
               Serge J. Belongie and
               Lubomir D. Bourdev and
               Ross B. Girshick and
               James Hays and
               Pietro Perona and
               Deva Ramanan and
               Piotr Doll{'{a} }r and
               C. Lawrence Zitnick},
  title     = {Microsoft {COCO:} Common Objects in Context},
  journal   = {CoRR},
  volume    = {abs/1405.0312},
  year      = {2014},
  url       = {},
  archivePrefix = {arXiv},
  eprint    = {1405.0312},
  timestamp = {Mon, 13 Aug 2018 16:48:13 +0200},
  biburl    = {},
  bibsource = {dblp computer science bibliography,}
  author    = {Andrej Karpathy and
               Fei{-}Fei Li},
  title     = {Deep visual-semantic alignments for generating image
  booktitle = { {IEEE} Conference on Computer Vision and Pattern Recognition,
               {CVPR} 2015, Boston, MA, USA, June 7-12, 2015},
  pages     = {3128--3137},
  publisher = { {IEEE} Computer Society},
  year      = {2015},
  url       = {},
  doi       = {10.1109/CVPR.2015.7298932},
  timestamp = {Wed, 16 Oct 2019 14:14:50 +0200},
  biburl    = {},
  bibsource = {dblp computer science bibliography,}

coco_captions/2014 (default config)