JettyCoffee/LiveCompose-outpainted-17k

---
pretty_name: LiveCompose Outpainted 17k (Tar Shards)
size_categories:
- 10K<n<100K
task_categories:
- image-classification
- image-to-image
- object-detection
tags:
- outpainting
- synthetic-data
- layout
- tar-shards
license: mit
---

# LiveCompose Outpainted 17k (Tar Shards)

## Dataset Summary

This is a tar-sharded release of the LiveCompose outpainted image dataset.
It is prepared for hosting on Hugging Face without hitting the per-directory file limit.

Contents:
- `training_pairs.jsonl` metadata
- `outpainted_tar/outpainted_part_001.tar` ... `outpainted_part_006.tar`

## Why Tar Shards

Hugging Face dataset repos currently enforce a limit on file count per directory.
The original `outpainted/` directory contains too many image files for a direct upload.
This release packs images into 6 tar shards.

## File Layout

```text
livecompose_outpainted_release/
README.md
training_pairs.jsonl
outpainted_tar/
outpainted_part_001.tar
outpainted_part_002.tar
outpainted_part_003.tar
outpainted_part_004.tar
outpainted_part_005.tar
outpainted_part_006.tar
```

## Statistics

- Original images: 17,684 PNG files
- Tar shards: 6
- Tar total size: about 12.615 GB

## Metadata Format

Each line in `training_pairs.jsonl`:

```json
{"file": "./outpainted/HORI2_000001_v01.png", "orig_bbox": [281, 77, 881, 487]}
```

Fields:
- `file`: relative path in the original unpacked layout
- `orig_bbox`: `[x_min, y_min, x_max, y_max]`

## How To Use

1. Download dataset files.
2. Extract tar shards into an `outpainted/` folder.
3. Read `training_pairs.jsonl` and resolve each `file` path.

Example extraction command (PowerShell):

```powershell
New-Item -ItemType Directory -Force -Path .\outpainted | Out-Null
Get-ChildItem .\outpainted_tar\*.tar | ForEach-Object {
tar -xf $_.FullName -C .\outpainted
}
```

## License

Current placeholder:
- `license: mit`

Please update this field before public release to match your redistribution rights.

## Citation

```bibtex
@dataset{livecompose_outpainted_17k_tar,
title={LiveCompose Outpainted 17k (Tar Shards)},
author={LiveCompose Team},
year={2026},
publisher={Hugging Face},
url={https://huggingface.co/datasets/JettyCoffee/LiveCompose-outpainted-17k}
}
```