PawChrona-MMQA contains 49,103 English–Chinese questions about companion-animal health and behavior, with images, audio, video and text. Cat and dog views are available through the species field.

PawChrona-MMQA 包含 49,103 道中英双语题目，涵盖宠物健康与行为的图像、音频、视频及文本；可按物种分别筛选猫和狗。

External evaluation leaderboard / 外部评测榜单

Data and splits / 数据与分区

Four configurations / 四个配置：image_vqa、audio_video_qa、text_qa、related_holdout.

Combine the first three configurations for 44,704 train and 2,500 dev questions. 1,899 questions are held separately. Dev has five topics, 500 questions each; topic labels and answer formats are fields within the existing configurations.

合并前三个配置可取得 44,704 道 train 和 2,500 道 dev；另有 1,899 道题单独暂缓。Dev 五大类各 500 题；内容类别及答题形式保存在现有配置的字段中。

Topic / 内容类别	Train	Dev	Held / 暂缓
Behavior and expression / 行为与表情	19,200	500	313
Disease categories and veterinary knowledge / 疾病类别与兽医知识	14,251	500	295
Imaging basics and quality / 成像基础与质量	2,209	500	304
Image observation and interpretation / 影像观察与判读	8,711	500	968
Pet health consultation / 宠物健康咨询	333	500	19

Format / 答题形式	Train	Dev	Held / 暂缓
`mcq4`	33,213	1,198	1,572
`mcq3`	1,082	292	230
`mcq2`	8,027	35	7
`open`	2,382	975	90

Usage / 使用

from datasets import load_dataset, concatenate_datasets
repo = "PawChrona/PawChrona-MMQA"
dev = concatenate_datasets([
    load_dataset(repo, name=name, split="dev", streaming=True, token=True)
    for name in ["image_vqa", "audio_video_qa", "text_qa"]
])
mcq4 = dev.filter(lambda x: x["question_format"] == "mcq4")
dogs = dev.filter(lambda x: x["species"] == "dog")

Use MCQ4 for four-choice accuracy; score MCQ2/3 and open responses separately. Pin a commit or release tag for comparisons. Exclude answers, explanations, evidence and concept IDs from model inputs.

四选一准确率使用 MCQ4；MCQ2/3 与开放题分别评分。比较成绩时固定提交或版本标签；模型输入排除答案、解释、证据及概念 ID。

Evaluation / 评测 · Fields / 字段 · Revision / 修订 · Train / 训练 · Dev / 开发

Sources and access / 来源与访问

Sources include veterinary paper figures, pet image collections, animal audio/video datasets and veterinary QA. Source links and credits are retained in metadata and third-party notices. This is an unreviewed research preview. Research access is reviewed individually; source terms remain applicable.

来源包括兽医论文图表、宠物图像、动物音视频及兽医问答；来源链接与署名保存在元数据和第三方来源说明中。本版为未经兽医审核的科研预览；科研访问逐人审核，保留来源原有条款。

Research access / 科研访问 · Commercial use / 商业使用 · Rights / 权利说明

Citation / 引用

PawChrona Team · 2026.10.30-debug.6

@misc{pawchrona2026mmqa,
  author = {{PawChrona Team}},
  title = {PawChrona-MMQA: Bilingual Companion-Animal Multimodal Question Answering},
  year = {2026},
  version = {2026.10.30-debug.6},
  url = {https://huggingface.co/datasets/PawChrona/PawChrona-MMQA}
}

Downloads last month: 120

Total file size:

41.7 GB