---
pretty_name: NCS-STR
task_categories:
- visual-question-answering
- image-to-text
language:
- en
configs:
- config_name: hands_on
  data_files:
  - split: train
    path: data/hands_on/*.parquet
- config_name: advanced
  data_files:
  - split: train
    path: data/advanced/*.parquet
---

# NCS-STR

NCS-STR contains two subsets:

- `hands_on`
- `advanced`

Each example contains a shuffled image sequence, an annotation, and the correct order.

## Usage after uploading to Hugging Face

```python
from datasets import load_dataset

hands_on = load_dataset("WM33333/NCS-STR", "hands_on", split="train", token=True)
advanced = load_dataset("WM33333/NCS-STR", "advanced", split="train", token=True)
```

## Fields
- `video_id`: folder id extracted from the original `vid` path
- `annotation`: task annotation
- `image_filenames`: shuffled frame filenames
- `images`: shuffled image sequence, stored as image bytes in Parquet
- `answer_order`: correct order, preserved exactly from the JSON files
- `subset`: `hands_on` or `advanced`