---
license: cc-by-4.0
task_categories:
- object-detection
- image-classification
pretty_name: PuzzleMapDS
size_categories:
- 1M<n<10M
---

# Puzzle-Map Dataset

The **Puzzle-Map Dataset** is a collection of images and annotations for research on computer vision applied to jigsaw puzzles.

## Directory Structure

```text
masks/
    *.png | *.jpg

masks-raw/
    *.png | *.jpg

pieces/
    *.png | *.jpg
    annotations.json

puzzles/
    *.png | *.jpg
    annotations.json
```

## Description

* **masks/** – Processed masks used for data augmentation.
* **masks-raw/** – Original masks used to generate the processed masks.
* **pieces/** – Cropped puzzle piece images and their annotations.
* **puzzles/** – Complete puzzle images and their annotations.

> **Note:** This dataset does **not** currently provide segmentation annotations. The mask images are auxiliary data used for data augmentation.

## Piece Annotations

Each image in the `pieces/` directory may contain one or more annotated puzzle pieces. The annotations are stored in `annotations.json`.

Each piece includes the following information:

* **Bounding box**
* **Image quality** (`valid`)
* **Piece rotation**
* **Side classification** (`INNER`, `OUTER`, `SMOOTH`)
* **Puzzle identification**
* **Puzzle row and column**
* **Piece angle inside the original puzzle**

Example:

```json
{
  "pieces": [
    {
      "bbox": [xmin, ymin, xmax, ymax],
      "valid": true,
      "rotation": 90,
      "sides": {
        "top": "OUTER",
        "right": "INNER",
        "bottom": "SMOOTH",
        "left": "INNER"
      },
      "puzzle": {
        "name": "puzzle_001",
        "row": 3,
        "column": 5,
        "angle": 180
      }
    }
  ]
}
```

## Acknowledgements

Part of the images and bounding box annotations included in the `pieces/` directory were built upon publicly available datasets from **Roboflow Universe**.

The original authors deserve full credit for collecting and publishing these datasets.

During the development of the Puzzle-Map Dataset, these original annotations were **extended** with additional information, including:

* Image quality labels (`valid`)
* Piece rotation
* Piece side classification (`INNER`, `OUTER`, `SMOOTH`)
* Puzzle identification
* Puzzle position (row and column)
* Piece orientation angle

These additional annotations were created specifically for the Puzzle-Map project.

## Original Datasets

### Puzzle Pieces Detection

* https://universe.roboflow.com/petro-telishevskyi-ewaoc/puzzle-pieces-detection

```bibtex
@misc{puzzle-pieces-detection_dataset,
  title = {Puzzle pieces detection Dataset},
  type = {Open Source Dataset},
  author = {Petro Telishevskyi},
  howpublished = {\url{https://universe.roboflow.com/petro-telishevskyi-ewaoc/puzzle-pieces-detection}},
  url = {https://universe.roboflow.com/petro-telishevskyi-ewaoc/puzzle-pieces-detection},
  journal = {Roboflow Universe},
  publisher = {Roboflow},
  year = {2023},
  month = {aug}
}
```

### Puzzle Dataset

* https://universe.roboflow.com/class-i1e8t/puzzle-vrkx6

```bibtex
@misc{puzzle-vrkx6_dataset,
  title = {puzzle Dataset},
  type = {Open Source Dataset},
  author = {class},
  howpublished = {\url{https://universe.roboflow.com/class-i1e8t/puzzle-vrkx6}},
  url = {https://universe.roboflow.com/class-i1e8t/puzzle-vrkx6},
  journal = {Roboflow Universe},
  publisher = {Roboflow},
  year = {2022},
  month = {aug}
}
```

## Citation

If you use this dataset in your research, please cite both this repository and the original Roboflow Universe datasets listed above.