jade-kai/handwriting-signatures-dataset

---
license: cc-by-nc-4.0
task_categories:
- text-to-image
- image-classification
language:
- und
tags:
- handwriting
- signatures
- lora
- diffusion
- computer-vision
- image-classification
- text-to-image
- stable-diffusion
pretty_name: Handwritten Signatures Dataset (Processed)
size_categories:
- 1K<n<10K
---

A dataset of handwritten signatures with text prompts, designed for LoRA fine-tuning of diffusion models to generate realistic personal signatures.

# Handwritten Signatures Dataset (Processed)

This dataset contains preprocessed handwritten signatures designed for **signature verification** and **LoRA fine-tuning** of diffusion models (e.g., Stable Diffusion) for text-to-image tasks.

## 📌 Processing steps

1. **Labeling** – assigned using computer vision and AI models to group signatures by name.
2. **Preprocessing with OpenCV** – cleaning, cropping, centering, and resizing to `256×256`.
3. **Manual filtering** – removal of low-quality or noisy samples.

## 📂 Structure

- `.png` files — signature images.
- `metadata.jsonl` — annotation file mapping each image to its label.

## 🚀 Usage

Example loading with `datasets`:

```python
from datasets import load_dataset

ds = load_dataset("sl4shuur/handwriting-signatures-dataset", split="train")
print(ds[0])
```

Each sample contains:

- `image`: PIL image object (signature image, 256×256).
- `text`: text prompt describing the signature (e.g., "signaturegen style: Close-up of B. Navya's handwritten signature in black ink on white").

## 🔖 Notes

* Original raw data: [Kaggle Signature Verification Dataset](https://www.kaggle.com/datasets/akashgundu/signature-verification-dataset).
* This version is heavily **processed and cleaned** for research and generative tasks.
* License is set to **CC-BY-NC-4.0** (non-commercial use only).

---