---
license: other
task_categories:
- visual-question-answering
- object-detection
language:
- vi
tags:
- vlm
- person-detection
- social-interaction
pretty_name: Vietnamese Social Interaction VLM Dataset
---

# Vietnamese Social Interaction VLM Dataset

The `train` directory follows the Hugging Face ImageFolder format. Each row has:

- `image`: JPEG image with person bounding boxes and IDs drawn on it.
- `ground_truth`: JSON string containing person boxes, states, and pair interaction labels.