| license: cc-by-4.0 | |
| tags: | |
| - photos | |
| - images | |
| - exif | |
| - photo-management | |
| size_categories: | |
| - 10K<n<50K | |
| # Photo Corpus | |
| A corpus of ~32k real-world photos with EXIF metadata, for developing and testing photo management software. | |
| ## Sources | |
| | Folder | Count | Source | EXIF date coverage | | |
| |--------|-------|--------|--------------------| | |
| | `yfcc/` | 31,385 | YFCC100M (Flickr originals) | ~78% | | |
| | `oi/` | 1,034 | Open Images (Flickr originals) | ~92% | | |
| Photos are Flickr originals with intact EXIF (GPS, dates, camera info) where available. Photos without EXIF DateTimeOriginal have file mtime set to the Flickr upload date as fallback. | |
| ## Format | |
| WebDataset-style tar shards (~1GB each): | |
| - `yfcc/yfcc-0000.tar` … `yfcc-0024.tar` | |
| - `oi/oi-0000.tar` … `oi-0002.tar` | |
| Each tar contains JPEG files named by content hash. | |
| ## Usage | |
| ```python | |
| import tarfile | |
| from pathlib import Path | |
| for shard in sorted(Path("yfcc").glob("*.tar")): | |
| with tarfile.open(shard) as tar: | |
| tar.extractall("photos/") | |
| ``` | |