Open Images is a dataset of ~9M images that have been annotated with image-level labels and object bounding boxes.

The training set of V4 contains 14.6M bounding boxes for 600 object classes on 1.74M images, making it the largest existing dataset with object location annotations. The boxes have been largely manually drawn by professional annotators to ensure accuracy and consistency. The images are very diverse and often contain complex scenes with several objects (8.4 per image on average). Moreover, the dataset is annotated with image-level labels spanning thousands of classes.

Split Examples
'test' 125,436
'train' 1,743,042
'validation' 41,620
  • Feature structure:
    'bobjects': Sequence({
        'bbox': BBoxFeature(shape=(4,), dtype=tf.float32),
        'is_depiction': tf.int8,
        'is_group_of': tf.int8,
        'is_inside': tf.int8,
        'is_occluded': tf.int8,
        'is_truncated': tf.int8,
        'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=601),
        'source': ClassLabel(shape=(), dtype=tf.int64, num_classes=6),
    'image': Image(shape=(None, None, 3), dtype=tf.uint8),
    'image/filename': Text(shape=(), dtype=tf.string),
    'objects': Sequence({
        'confidence': tf.int32,
        'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=19995),
        'source': ClassLabel(shape=(), dtype=tf.int64, num_classes=6),
    'objects_trainable': Sequence({
        'confidence': tf.int32,
        'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=7186),
        'source': ClassLabel(shape=(), dtype=tf.int64, num_classes=6),
  • Feature documentation:
Feature Class Shape Dtype Description
bobjects Sequence
bobjects/bbox BBoxFeature (4,) tf.float32
bobjects/is_depiction Tensor tf.int8
bobjects/is_group_of Tensor tf.int8
bobjects/is_inside Tensor tf.int8
bobjects/is_occluded Tensor tf.int8
bobjects/is_truncated Tensor tf.int8
bobjects/label ClassLabel tf.int64
bobjects/source ClassLabel tf.int64
image Image (None, None, 3) tf.uint8
image/filename Text tf.string
objects Sequence
objects/confidence Tensor tf.int32
objects/label ClassLabel tf.int64
objects/source ClassLabel tf.int64
objects_trainable Sequence
objects_trainable/confidence Tensor tf.int32
objects_trainable/label ClassLabel tf.int64
objects_trainable/source ClassLabel tf.int64
  author = {Alina Kuznetsova and
            Hassan Rom and
            Neil Alldrin and
            Jasper Uijlings and
            Ivan Krasin and
            Jordi Pont-Tuset and
            Shahab Kamali and
            Stefan Popov and
            Matteo Malloci and
            Tom Duerig and
            Vittorio Ferrari},
  title = {The Open Images Dataset V4: Unified image classification,
           object detection, and visual relationship detection at scale},
  year = {2018},
  journal = {arXiv:1811.00982}
  author = {Krasin, Ivan and
            Duerig, Tom and
            Alldrin, Neil and
            Ferrari, Vittorio
            and Abu-El-Haija, Sami and
            Kuznetsova, Alina and
            Rom, Hassan and
            Uijlings, Jasper and
            Popov, Stefan and
            Kamali, Shahab and
            Malloci, Matteo and
            Pont-Tuset, Jordi and
            Veit, Andreas and
            Belongie, Serge and
            Gomes, Victor and
            Gupta, Abhinav and
            Sun, Chen and
            Chechik, Gal and
            Cai, David and
            Feng, Zheyun and
            Narayanan, Dhyanesh and
            Murphy, Kevin},
  title = {OpenImages: A public dataset for large-scale multi-label and
           multi-class image classification.},
  journal = {Dataset available from

open_images_v4/original (default config)

  • Config description: Images at their original resolution and quality.


  • Config description: Images have roughly 300,000 pixels, at 72 JPEG quality.


  • Config description: Images have roughly 200,000 pixels, at 72 JPEG quality.