Datasets:

kg-09
/

SkinTone-Dataset

license: mit
task_categories:
  - image-to-image
  - image-classification
tags:
  - skin-tone-matching
  - inpainting
  - lora-training
  - computer-vision
pretty_name: Skin Tone Matching LoRA Training Dataset
size_categories:
  - n<1K

Skin Tone Matching LoRA Training Dataset

Dataset Description

This dataset is designed for training LoRA (Low-Rank Adaptation) models for skin tone matching and inpainting tasks. The dataset contains 98 image pairs with corresponding text prompts for training generative models to perform natural-looking skin inpainting that matches surrounding skin tone, texture, and lighting.

Dataset Structure

The dataset contains the following fields:

image_id: Unique identifier for each image pair
input_image: Input image with green masked areas that need to be inpainted
target_image: Ground truth target image showing the desired inpainting result
caption: Text prompt describing the inpainting task
task: Task category (always "skin_tone_matching")

Use Cases

This dataset is suitable for:

Training LoRA adapters for skin tone matching
Fine-tuning diffusion models for inpainting tasks
Research in computer vision and image generation
Developing tools for photo editing and restoration

Dataset Creation

The dataset was created for training models to:

Identify green masked areas in facial images
Generate natural-looking skin that matches the surrounding area
Ensure seamless blending with existing skin tone, texture, and lighting
Maintain realistic shadows and highlights

Citation

If you use this dataset in your research, please cite:

@dataset{skin_tone_matching_lora,
  title={Skin Tone Matching LoRA Training Dataset},
  author={Karthik},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/kg-09/SkinTone-Dataset}
}

License

This dataset is released under the License.