• Description:

The PatchCamelyon benchmark is a new and challenging image classification dataset. It consists of 327.680 color images (96 x 96px) extracted from histopathologic scans of lymph node sections. Each image is annoted with a binary label indicating presence of metastatic tissue. PCam provides a new benchmark for machine learning models: bigger than CIFAR10, smaller than Imagenet, trainable on a single GPU.

Split Examples
'test' 32,768
'train' 262,144
'validation' 32,768
  • Feature structure:
    'id': Text(shape=(), dtype=string),
    'image': Image(shape=(96, 96, 3), dtype=uint8),
    'label': ClassLabel(shape=(), dtype=int64, num_classes=2),
  • Feature documentation:
Feature Class Shape Dtype Description
id Text string
image Image (96, 96, 3) uint8
label ClassLabel int64


  • Citation:
  author       = {B. S. Veeling, J. Linmans, J. Winkens, T. Cohen, M. Welling},
  title        = {Rotation Equivariant CNNs for Digital Pathology},
  month        = sep,
  year         = 2018,
  doi          = {10.1007/978-3-030-00934-2_24},
  url          = {}