oxford_iiit_pet

• Description:

The Oxford-IIIT pet dataset is a 37 category pet image dataset with roughly 200 images for each class. The images have large variations in scale, pose and lighting. All images have an associated ground truth annotation of breed and species. Additionally, head bounding boxes are provided for the training split, allowing using this dataset for simple object detection tasks. In the test split, the bounding boxes are empty.

• Additional Documentation: Explore on Papers With Code north_east

• Homepage: http://www.robots.ox.ac.uk/~vgg/data/pets/

• Source code: tfds.datasets.oxford_iiit_pet.Builder

• Versions:

  • 4.0.0 (default): Add head bounding boxes. Fix corrupt images. Update dataset URL.

• Download size: 773.52 MiB

• Dataset size: 773.68 MiB

• Auto-cached (documentation): No

• Splits:

• Feature structure:

FeaturesDict({
    'file_name': Text(shape=(), dtype=string),
    'head_bbox': BBoxFeature(shape=(4,), dtype=float32),
    'image': Image(shape=(None, None, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=37),
    'segmentation_mask': Image(shape=(None, None, 1), dtype=uint8),
    'species': ClassLabel(shape=(), dtype=int64, num_classes=2),
})

• Feature documentation:

• Supervised keys (See as_supervised doc): ('image', 'label')

• Figure (tfds.show_examples): Not supported.

• Examples (tfds.as_dataframe):

• Citation:

@InProceedings{parkhi12a,
  author       = "Parkhi, O. M. and Vedaldi, A. and Zisserman, A. and Jawahar, C.~V.",
  title        = "Cats and Dogs",
  booktitle    = "IEEE Conference on Computer Vision and Pattern Recognition",
  year         = "2012",
}