VerboVision/Captions-5K
收藏Hugging Face2026-03-27 更新2026-03-29 收录
下载链接：
https://hf-mirror.com/datasets/VerboVision/Captions-5K
下载链接
链接失效反馈官方服务：
资源简介：
---
dataset_info:
features:
- name: image_id
dtype: int64
- name: image
dtype: image
- name: caption
dtype: string
- name: source
dtype: string
- name: caption_gemini-2.5-pro
dtype: string
- name: caption_gemini-3.1-flash-lite-preview
dtype: string
- name: caption_gemini-2.5-flash
dtype: string
- name: caption_vllm_qwen35_detail
dtype: string
- name: caption_vllm_qwen35_pretrained
dtype: string
- name: caption_vllm_qwen35_27b
dtype: string
- name: caption_aya_vision_8b
dtype: string
- name: caption_qwen3vl_4b
dtype: string
splits:
- name: train
num_bytes: 572815308
num_examples: 5000
download_size: 551933145
dataset_size: 572815308
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
# Dataset de captions geradas por:
## Três modelos Gemini -> gemini-2.5-pro // gemini-3.1-flash-lite-preview // gemini-2.5-flash
## Três modelos Open Source -> Qwen3-VL-4B-Instruct // aya_vision_8b // Qwen3.5-27B
## Dois modelos VerboVision -> qwen3.5-4b-pretrained-merged // qwen3.5-4b-verbovision-detail-tags-merged
## -----------------------------------------------------------------------------------------------------------
# Dataset base: VerboVision/PraCegoVer-Filtrado-FSB
## -----------------------------------------------------------------------------------------------------------
# Quantidade de amostras:
## gemini-2.5-pro -> ~1000 amostras
## Outros -> ~5000 amostras
## -----------------------------------------------------------------------------------------------------------
# Dataset base para seleção por preferência para treinamento por reforço
提供机构：
VerboVision
