---
language:
- en
pretty_name: FFHQ Proxy Reference
size_categories:
- 10K<n<100K
tags:
- image-restoration
- portrait-upscaling
- face-super-resolution
- aesthetic-refinement
license: creativeml-openrail-m
task_categories:
- image-to-image
---

# FFHQ Proxy Reference

## Overview

**FFHQ Proxy Reference** is a **reference-only dataset repository** designed to integrate high-quality human face images into advanced image restoration and portrait upscaling pipelines.

This repository acts as a **proxy schema and access reference** for the official FFHQ dataset and is optimized for:

- Facial detail preservation
- Portrait-focused super-resolution (up to 8K)
- Identity-consistent restoration workflows

> **Important**  
> This repository does **not host image files**.  
> All images must be obtained directly from the official FFHQ source and loaded locally or via private storage using a custom dataset loader.

---

## Intended Use

This dataset reference is designed for:

- Portrait upscaling models
- Face restoration and enhancement pipelines
- Aesthetic refinement workflows
- High-resolution facial reconstruction tasks

It is **not** intended to redistribute FFHQ images or bypass licensing restrictions.

---

## Dataset Structure (Expected)

Each sample is expected to follow this structure when loaded locally:

```
| Field | Type | Description |
|------|------|------------|
| `image` | `Image` | High-quality face image |
| `resolution` | `int` | Image resolution indicator |
| `source` | `string` | Original dataset reference |
```
---

## Data Source

- **Primary Source (Official FFHQ):**  
https://github.com/NVlabs/ffhq-dataset

- **Mirror / Access Reference:**
  Google Drive (official distribution channel).

All images remain under the **original FFHQ license and terms**.

---

## Integration Notes

- Hugging Face **does not ingest Google Drive folders directly**
- This repository serves as:
  - A dataset reference
  - A schema definition
  - A controlled integration point for private pipelines

To use the data, users must:
1. Download FFHQ locally from the official source
2. Load images using a custom `dataset.py` or local loader
3. Ensure compliance with FFHQ licensing

---

## Example Usage

```python
from datasets import load_dataset

dataset = load_dataset(
    "Legitking4pf/FFHQ-Proxy-Reference",
    data_dir="/path/to/local/ffhq"
)
```

## Licensing 

- **Repository contents:** CreativeML Open RAIL-M.

- **Image data:** Governed entirely by the original FFHQ license.

This repository does not grant redistribution rights for FFHQ images.

## Disclaimer

This dataset is a proxy reference only. It exists to standardize integration and does not replace or mirror the original dataset.

- **Users are responsible for:**
  - Data access
  - Storage
  - Licensing compliance
  - Copy code

---