13point5/line-ex

---
pretty_name: LineEX
task_categories:
- image-classification
- object-detection
- image-to-text
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
---

# LineEX

This dataset repo contains:

- the uploaded `train` split with `397,993` images
- the released `test` split with `20,000` images

- Repo: `13point5/line-ex`

## Shared Schema

- `image_id`
- `file_name`
- `image`
- `width`
- `height`
- `data_type`
- `chart_elements`
- `lines`

## `chart_elements` fields

- `annotation_id`
- `category_id`
- `category_name`
- `bbox_xywh`
- `area`
- `text`
- `line_id`

## `lines` fields

- `annotation_id`
- `category_id`
- `category_name`
- `line_name`
- `polyline_xy`
- `raw_series_xy`
- `area`

## Notes

- The repo uses a shared nested schema across splits.
- The `train` split is streamed directly from `train.zip` and uploaded as parquet shards to
reduce local disk usage.
- The `test` split uses the 10-class chart-element mapping defined in the LineEX codebase.
- The `train` split uses a different chart-element label vocabulary, but the HF feature shape
is identical.
- `lines.line_name` is derived from legend structure and may be `null` when the legend-text
association is ambiguous.

## License

This mirror does not assert a new license for the data. Verify redistribution terms with the
original dataset authors before making the dataset public.