configs/dataset/coco2014_20k.yaml

# The COCO20k dataset with instance and caption annotations.
#
# COCO 20k is a subset of COCO 2014. It is a random subset of both the training
# and validation sets, filtered to not contain any images only having crowd instances
# and no annotations labeled as crowd. It is commonly used to evaluate object
# discovery methods.
#
# This configuration defines COCO20k as a validation set.
_target_: ocl.datasets.WebdatasetDataModule
val_shards: ${oc.env:DATASET_PREFIX}/coco2014/20k/shard-{000000..000064}.tar
val_size: 19820
use_autopadding: true