---
dataset_info:
  features:
  - name: idx
    dtype: int64
  - name: video_url
    dtype: string
  - name: question_category
    dtype: string
  - name: question
    dtype: string
  - name: options
    struct:
    - name: A
      dtype: string
    - name: B
      dtype: string
    - name: C
      dtype: string
    - name: D
      dtype: string
    - name: E
      dtype: string
  - name: correct_answer
    dtype: string
  splits:
  - name: test
    num_bytes: 3577381
    num_examples: 4744
  download_size: 1830493
  dataset_size: 3577381
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
license: cc-by-nc-4.0
task_categories:
  - visual-question-answering
language:
  - en
tags:
  - dogs
  - canine-centric
  - video-question-answering
  - benchmark
  - multiple-choice
pretty_name: K9Bench
size_categories:
  - 1K<n<10K
extra_gated_prompt: "Terms of Use\n\nFor Your Information\nThe dataset provides access to videos, images, metadata, and/or links to videos obtained from YouTube and other publicly available sources. The dataset authors, maintainers, and affiliated organizations do not own the copyrights to the videos or other third-party content referenced or linked within the dataset.\nOwnership and copyright of all videos and related content remain with the respective YouTube channel owners, creators, or rights holders. It is the responsibility of the original source channels and content owners to ensure that their content complies with YouTube's terms, policies, and applicable laws.\nBy accessing or using this dataset, you acknowledge and agree to the following terms:\n\n1. Permitted Use\nThe dataset may be used only for non-commercial scientific, academic, or research purposes.\nAny commercial use, including but not limited to product development, commercial training, resale, licensing, redistribution, or use in commercial services, is strictly prohibited unless prior written permission is obtained from the dataset authors.\n\n2. Copyright and Ownership\nThe dataset authors do not claim ownership of the YouTube videos or any third-party media referenced in the dataset.\nYou understand and acknowledge that:\n- The dataset may contain links to YouTube videos or references to publicly available media.\n- The dataset authors do not own the videos linked or referenced in the dataset.\n- Copyright and ownership of the videos remain with the respective content owners.\n- You are responsible for ensuring that your use of any linked or referenced content complies with applicable copyright laws, YouTube's Terms of Service, and any other relevant licenses or restrictions.\n\n3. Restrictions on Sharing and Redistribution\nYou agree that the dataset must not be shared, distributed, published, transferred, sublicensed, or redistributed, in whole or in part, to any third-party individual, group, institution, company, or organization.\nAccess to the dataset is granted only to the approved data requestor or authorized user.\n\n4. Use of Videos and Images in Publications\nVideos, images, frames, or visual examples from the dataset may only be shown in academic publications, research papers, theses, posters, or academic presentations.\nAny such use must be limited to what is necessary for scholarly communication and must properly acknowledge the dataset and, where applicable, the original content source.\n\n5. Compliance with Dataset License\nBy accessing or using the dataset, you agree to comply with the applicable dataset license and all terms stated in this Terms of Use document.\nYou specifically acknowledge and agree that:\n- You understand that the dataset authors do not own the YouTube videos linked or referenced in the dataset;\n- You will use the dataset for non-commercial research purposes only;\n- You will not redistribute or share the dataset with unauthorized third parties;\n- You will not create a new or derivative dataset without written consent;\n- You agree to the dataset license and all usage restrictions.\n\n6. Termination of Access\nThe dataset authors reserve the right to revoke or terminate access to the dataset if you violate any of these Terms of Use or use the dataset in a manner inconsistent with the stated research and non-commercial purposes.\n\n7. No Warranty\nThe dataset is provided 'as is' without warranties of any kind, express or implied. The dataset authors make no guarantees regarding the availability, accuracy, legality, completeness, or continued accessibility of any linked videos or third-party content.\n\n8. Acceptance\nBy downloading, accessing, or using the dataset, you confirm that you have read, understood, and agree to be bound by these Terms of Use."
extra_gated_fields:
  First Name: text
  Last Name: text
  Email: text
  Affiliation: text
  Country: country
  I understand that K9Bench does not own the YouTube videos linked in this dataset: checkbox
  I agree to use this dataset for non-commercial use ONLY: checkbox
  I agree with the data license for this dataset: checkbox
extra_gated_button_content: "Agree and access repository"
---

# K9Bench 🐕

K9Bench is a video question-answering benchmark for evaluating fine-grained understanding of canine actions, and interactions in real-world YouTube videos.


## Dataset Summary

Each sample pairs a YouTube video with a multiple-choice question requiring detailed visual understanding of dog behavior. Questions span five categories: action sequence, cause-effect analysis, context analysis, interaction analysis, and posture analysis.

## Dataset Structure

| Field | Type | Description |
|---|---|---|
| `idx` | int | Unique sequential index |
| `video_url` | string | Full YouTube URL |
| `question_category` | string | One of 5 categories |
| `question` | string | The question about the video |
| `options` | dict | Five answer choices (A–E) |
| `correct_answer` | string | Ground-truth answer (A–E) |

## Question Categories

- **action sequence** — ordering of behavioral events
- **cause-effect analysis** — causal relationships between actions
- **context analysis** — situational and environmental influences
- **interaction analysis** — social signals between animals or humans
- **posture analysis** — body language and physical cues

## Splits

| Split | Rows |
|---|---|
| test | 4,744 |

## Usage

```python
from datasets import load_dataset

ds = load_dataset("ogmen/K9Bench")
sample = ds["test"][0]
print(sample["question"])
print(sample["options"])
print(sample["correct_answer"])
```

## Citation

```bibtex
@dataset{k9bench2026,
  title   = {K9-Bench: Evaluating Multimodal LLMs on Canine-Centric Videos},
  year    = {2026},
  url     = {https://huggingface.co/datasets/ogmen/K9Bench}
}
```