TierKun/Indonesian-running-photos

Dataset Card for Fotoyu Album Archive

This dataset stores photo and video collections archived from Fotoyu albums using the potoyu-tree-downloader application. It is designed to act as a high-speed Cloudflare-backed CDN for serving static media assets, as well as providing a dataset for image/video classification and machine learning model training.

Dataset Details

Dataset Description

The dataset aggregates scraped photo galleries and video albums (stored as HLS stream directories) and serves them globally via Hugging Face's direct resolution endpoints. All files are organized and sharded to guarantee compatibility with Hugging Face's directory file count limits.

• Curated by: System administrators using potoyu-tree-downloader
• Shared by: Admin contributors
• Language(s): English, Indonesian (for album labels/metadatas)
• License: Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)

Dataset Sources

• Repository: potoyu-tree-downloader GitHub

---

Uses

Direct Use

• Static Asset Serving (CDN): The primary purpose of this dataset is to serve as a public CDN. Frontend applications fetch optimized images and stream videos directly from the repository's resolve/main/ paths.
• Classification & Computer Vision: The dataset can be utilized to train and evaluate classification models on custom human activity, events, and running photography classifications.

Out-of-Scope Use

• This dataset should not be used for commercial advertising without explicit permission from the original photo owners and creators.

---

Dataset Structure

The repository implements a 2-character prefix sharding system to prevent folders from exceeding the 10,000 files/directory commit limit on Hugging Face.

Directory Layout

dataset-root/
└── {album-alias}/
    ├── {prefix}/                     # Last 2 chars of the unique leafId
    │   └── {leafId}.jpg              # Original image file or video thumbnail
    │
    └── videos/
        └── {prefix}/
            └── {leafId}/
                ├── index.m3u8        # HLS streaming playlist
                ├── segment0.ts       # Video stream chunks
                ├── segment1.ts
                └── ...

• {album-alias}: The customized alphanumeric name representing the specific gallery or album.
• {prefix}: The last two characters of the unique alphanumeric leafId (e.g., if leafId is ...a7b9c2, the prefix is c2). This partitions the files into up to 3,844 directories, ensuring directories never contain more than a few hundred files.
• videos/{prefix}/{leafId}/: Hosts HLS streaming playlist files (.m3u8) and segments (.ts) for video content.

---

Dataset Creation

Curation Rationale

Serving thousands of high-resolution images and videos directly from a single VPS results in high network bandwidth costs and latency for global users. Syncing assets to Hugging Face datasets offloads all media requests to Hugging Face's robust global edge cache (Cloudflare), resulting in faster load times and zero bandwidth charges.

Source Data

The source data is populated from Fotoyu trees (user-curated photo albums).

Data Collection and Processing

1. Scraping: The backend downloader connects to the Fotoyu API endpoints using a shared authorization bearer token.
2. Local Archiving: Media files are downloaded, structured, and saved locally. Video playlists are parsed and their segment files (.ts) are fully archived.
3. Sharding & Upload: The @huggingface/hub integration groups pending files into batches of 100 items. These are uploaded in a single git commit every 40 seconds to prevent hitting Hugging Face's 128 commits/hour rate limits.

---

Bias, Risks, and Limitations

• Rate Limits: Repository commits are capped at 128 per hour. Batch updates must remain throttled.
• Privacy: Archived public photos may contain personal identities. Appropriate privacy boundaries and licenses should be respected.