---
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 |
|-------|------|
| data_id | large_string |
| image_id | large_string |
| question | large_string |

## Splits

| Split | Rows | Size |
|-------|------|------|
| oven_entity_test | 709,478 | 1.5 MB |
| oven_entity_train | 4,926,314 | 8.9 MB |
| oven_entity_val | 126,199 | 0.5 MB |
| oven_query_test | 19,781 | 0.3 MB |
| oven_query_train | 32,255 | 0.5 MB |
| oven_query_val | 3,291 | 0.1 MB |
| **Total** | **5,817,318** | |

## Usage

```python
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")
```