arkimjh/werewolf2.0
收藏Hugging Face2026-03-27 更新2026-03-29 收录
下载链接:
https://hf-mirror.com/datasets/arkimjh/werewolf2.0
下载链接
链接失效反馈官方服务:
资源简介:
---
license: cc-by-4.0
task_categories:
- video-text-to-text
- visual-question-answering
language:
- en
tags:
- video
- social-interaction
- benchmark
- gaze
- werewolf
size_categories:
- 1K<n<10K
---
# Werewolf 2.0 Benchmark
## Description
Werewolf 2.0 is a video-based social reasoning benchmark dataset containing **2,487 video clips** with per-clip QA annotations. Each QA tests gaze-based social understanding in multi-person scenarios.
## Dataset Statistics
| Type | Count | Size |
|------|-------|------|
| Videos | 2,487 | ~848 MB |
| JSONs | 2,487 | ~11 MB |
## 16-Category QA Taxonomy
| Cat | Name | Difficulty | Format |
|-----|------|-----------|--------|
| **T1** | Gaze Target ID | Easy | MCQ |
| **T2** | Gaze Event Classification | Easy | MCQ |
| **T3** | Temporal Gaze Reasoning | Medium | MCQ |
| **T4** | Mutual Gaze Recognition | Medium | MCQ |
| **T5** | Gaze Following | Hard | MCQ |
| **T6** | Group Attention Dynamics | Hard | MCQ+OE |
| **G1** | Gesture Recognition | Easy | MCQ |
| **G2** | Gesture Type Classification | Easy | MCQ |
| **G3** | Gesture Temporal Reasoning | Medium | MCQ |
| **G4** | Gesture Frequency | Medium | MCQ |
| **G5** | Gesture Sequence Chains | Hard | MCQ+OE |
| **G6** | Reciprocal Gesture Patterns | Hard | MCQ |
| **C1** | Gaze-Gesture Temporal Alignment | Medium | MCQ |
| **C2** | Eye Contact During Interaction | Medium | MCQ |
| **C3** | Gaze Response to Gesture | Hard | MCQ+OE |
| **C4** | Cross-Modal Person Dynamics | Hard | MCQ+OE |
## Dataset Structure
```
werewolf2.0/
├── tars/
│ ├── werewolf2.0_videos_part001.tar → video/*.mp4
│ └── werewolf2.0_jsons_part001.tar → json/*.json
└── README.md
```
## Usage
```python
from huggingface_hub import snapshot_download
import tarfile
from pathlib import Path
# Download
snapshot_download(
repo_id="arkimjh/werewolf2.0",
repo_type="dataset",
local_dir="./werewolf2.0"
)
# Extract
for tar_file in Path("./werewolf2.0/tars").glob("*.tar"):
with tarfile.open(tar_file) as tf:
tf.extractall("./werewolf2.0/")
```
## Citation
TBD
提供机构:
arkimjh



