metadata
dataset_info:
features:
- name: id
dtype: int32
- name: image
dtype: image
- name: objects
list: string
- name: attributes
list:
- name: name
dtype: string
- name: attrs
list: string
- name: region_descriptions
list: string
- name: contrastive_attributes
list:
- name: attr1
dtype: string
- name: attr2
dtype: string
- name: nouns
list: string
splits:
- name: train
num_bytes: 15578774088
num_examples: 108077
download_size: 15573757911
dataset_size: 15578774088
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Visual Genome (108,077 images). Canonical shards live under data/.