metadata
dataset_info:
features:
- name: image
dtype: image
- name: image_name
dtype: string
- name: raw_response
dtype: string
- name: layout_caption
dtype: string
- name: overview_caption
dtype: string
- name: primary_object_caption
dtype: string
- name: secondary_object_caption
dtype: string
- name: interaction_caption
dtype: string
- name: parsed_to_json
dtype: bool
splits:
- name: train
num_bytes: 2064637569.423
num_examples: 11199
download_size: 1680472358
dataset_size: 2064637569.423
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
OpenViVQA Vietnamese Captions Dataset
Introduction
This dataset includes 11,199 images from the OpenViVQA collection, each paired with five Vietnamese captions automatically generated by the Qwen2.5-VL-32B model:
- layout_caption: Describes the overall layout of the image
- overview_caption: Provides a detailed overview (long caption)
- primary_object_caption: Describes the main object in detail (long caption)
- secondary_object_caption: Describes secondary objects
- interaction_caption: Describes interactions between objects