---
license: other
task_categories:
- image-segmentation
language:
- en
pretty_name: Curvilinear Data
size_categories:
- 1K<n<10K
---

# Curvilinear Data

This dataset contains curated curvilinear-structure segmentation samples for eight categories:

- `branch`
- `crack`
- `floor`
- `leaf`
- `scratch`
- `soil`
- `tyre`
- `wire`

The current curated split is stored under `1k_size_handle/`.

## Directory Layout

Each class has three directories:

- `<class>/`: RGB input images
- `<class>_label/`: binary segmentation masks
- `<class>_centerline_label/`: centerline masks

The dataset also includes:

- `1k_size_handle/image2label_train.json`
- `1k_size_handle/image2centerline_train.json`
- `1k_size_handle/renumber_summary.json`
- `dataset_summary.json`

`image2label_train.json` maps each image path to its segmentation mask path.
`image2centerline_train.json` maps each image path to its centerline mask path.

## Class Counts

| Class | Samples |
|---|---:|
| branch | 104 |
| crack | 102 |
| floor | 97 |
| leaf | 102 |
| scratch | 83 |
| soil | 89 |
| tyre | 118 |
| wire | 97 |

Total: 792 image-mask pairs.

## Notes

The files in `1k_size_handle/` have been filtered so that each retained label and centerline label has a matching source image. File names have been renumbered consecutively within each class.