README.md

metadata

dataset_info:
  features:
    - name: data_id
      dtype: string
    - name: image_id
      dtype: string
    - name: image
      dtype: string
    - name: question
      dtype: string
    - name: answer
      dtype: string
    - name: answer_eval
      dtype: string
    - name: entity_id
      dtype: string
    - name: entity_text
      dtype: string
    - name: attribute
      dtype: string
license: apache-2.0
task_categories:
  - visual-question-answering
  - information-retrieval
tags:
  - vision-self-play
  - multimodal
  - knowledge-intensive

OVEN (Vision Self-Play format)

Open-domain Visual Entity Recognition (OVEN), converted to unified Parquet schema.

Hu et al., Open-domain Visual Entity Recognition: Towards Recognizing Millions of Wikipedia Entities. ICCV 2023.

Schema

Field	Type	Description
data_id	string	Unique sample ID
image_id	string	Image identifier
image	string	Image path/URL (empty, resolve at training time)
question	string	Question text
answer	string (JSON list)	Standard answers
answer_eval	string (JSON list)	Acceptable answer variants
entity_id	string	Wikidata QID
entity_text	string	Entity name
attribute	string	Attribute type

Splits

Split	Rows	Size
test	729,259	1.8 MB
train	4,958,569	9.6 MB
val	129,490	0.6 MB
Total	5,817,318

Usage

from datasets import load_dataset

ds = load_dataset("reonokiy/vsp-oven")

# 或加载特定 split
train = load_dataset("reonokiy/vsp-oven", split="train")