Jharoon/lora-garment-textures

LoRA Garment Texture Training Dataset

<div align="center"> <img src="https://img.shields.io/badge/Task-LoRA%20Training-blue" alt="Task"> <img src="https://img.shields.io/badge/Domain-Fashion%20Textures-purple" alt="Domain"> <img src="https://img.shields.io/badge/Images-190-green" alt="Size"> <img src="https://img.shields.io/badge/License-MIT-yellow" alt="License"> </div>

📋 Dataset Description

This dataset contains high-quality garment texture images organized by categories for training LoRA (Low-Rank Adaptation) models. These images are specifically curated for fine-tuning diffusion models to generate virtual try-on results with specific fabric textures and patterns.

Key Features

• —🎨 Multiple texture categories for diverse garment styles
• —📸 High-resolution images suitable for LoRA training
• —🏷️ Organized by category for easy filtering and training
• —🎯 Fashion-focused textures for virtual try-on applications
• —🔧 Ready for training with metadata included

📊 Dataset Statistics

• —denim
• —linen
• —pattern1
• —pattern2

Total Images: 190

Image Properties

• —Formats: JPEG, PNG
• —Resolutions: Variable (high-resolution)
• —Color Mode: RGB
• —Categories: 4

🎯 Use Cases

1. LoRA Model Training

Train LoRA adapters for specific fabric textures:

from datasets import load_dataset

# Load dataset
dataset = load_dataset("zyuzuguldu/lora-garment-textures")

# Filter by category
denim_images = dataset["train"].filter(lambda x: x["category"] == "denim")

# Use in your LoRA training pipeline
for sample in denim_images:
    image = sample["image"]
    category = sample["category"]
    # Your training code here

2. Virtual Try-On Fine-tuning

Fine-tune diffusion models to generate specific garment textures in virtual try-on applications.

3. Texture-Conditioned Generation

Use as reference images for texture-aware garment generation models.

4. Style Transfer

Apply specific fabric textures to garment designs using style transfer techniques.

📁 Dataset Structure

Each sample contains:

• —`image_id` (string): Unique identifier (format: category_####)
• —`image` (PIL Image): The texture image
• —`category` (string): Texture category name
• —`filename` (string): Original filename
• —`width` (int): Image width in pixels
• —`height` (int): Image height in pixels
• —`mode` (string): Image color mode (e.g., "RGB")
• —`format` (string): Image file format (e.g., "JPEG", "PNG")

Example

{
    "image_id": "denim_0001",
    "image": <PIL.Image>,
    "category": "denim",
    "filename": "pexels-photo-7794357.jpeg",
    "width": 2048,
    "height": 1536,
    "mode": "RGB",
    "format": "JPEG"
}

🚀 Quick Start

Load Dataset

from datasets import load_dataset

# Load full dataset
dataset = load_dataset("zyuzuguldu/lora-garment-textures")

# Access the data
train_data = dataset["train"]

# Preview first sample
print(train_data[0])

Filter by Category

from datasets import load_dataset

dataset = load_dataset("zyuzuguldu/lora-garment-textures", split="train")

# Get all categories
categories = set(dataset["category"])
print(f"Available categories: {categories}")

# Filter specific category
denim_data = dataset.filter(lambda x: x["category"] == "denim")
print(f"Denim images: {len(denim_data)}")

Visualize Images

import matplotlib.pyplot as plt
from datasets import load_dataset

dataset = load_dataset("zyuzuguldu/lora-garment-textures", split="train")

# Display images from each category
fig, axes = plt.subplots(2, 2, figsize=(12, 12))

for idx, ax in enumerate(axes.flat):
    sample = dataset[idx]
    ax.imshow(sample["image"])
    ax.set_title(f"{sample['category']} - {sample['image_id']}")
    ax.axis("off")

plt.tight_layout()
plt.show()

LoRA Training Integration

from datasets import load_dataset
from diffusers import StableDiffusionPipeline, AutoencoderKL
from diffusers.loaders import AttnProcsLayers
import torch

# Load dataset
dataset = load_dataset("zyuzuguldu/lora-garment-textures", split="train")

# Filter for specific texture
texture_data = dataset.filter(lambda x: x["category"] == "denim")

# Load base model
model_id = "stabilityai/stable-diffusion-2-1"
pipe = StableDiffusionPipeline.from_pretrained(model_id)

# Your LoRA training code here
# Use texture_data as training input

🎨 Categories

The dataset includes the following texture categories:

• —denim
• —linen
• —pattern1
• —pattern2

Each category contains carefully curated images representing specific fabric textures or patterns commonly used in fashion and garment design.

💡 Best Practices

For LoRA Training

1. 1.Start with one category: Train separate LoRAs for each texture type
2. 2.Use consistent prompts: Include category name in your training prompts
3. 3.Adjust learning rates: Fine-tune for fabric texture details
4. 4.Monitor overfitting: Use validation images to check generalization

For Virtual Try-On

1. 1.Combine with segmentation: Use garment masks for targeted texture application
2. 2.Preserve garment structure: Maintain original garment shape while applying texture
3. 3.Match lighting: Ensure texture matches the lighting of the target image

📝 Dataset Creation

This dataset was curated from various sources for the specific purpose of training LoRA models for virtual try-on applications. Images were:

1. 1.Collected from high-quality fabric and texture sources
2. 2.Organized by texture category
3. 3.Verified for quality and resolution
4. 4.Prepared with metadata for easy filtering

⚖️ License

This dataset is released under the MIT License.

• —Free to use for commercial and non-commercial purposes
• —Attribution appreciated but not required

🔗 Related Resources

• —Garment Segmentation Model: garment-segmentation-unet-resnet50
• —Segmentation Dataset: deepfashion2-upper-body-masks
• —Gradio Demo: garment-segmentation

🤝 Contributing

Want to add more texture categories or improve the dataset? Feel free to reach out!

👨‍💻 Maintainer

Created and maintained by @zyuzuguldu

Made with ❤️ for the fashion-tech and AI community

📚 Citation

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

@dataset{lora_garment_textures,
  author = {zyuzuguldu},
  title = {LoRA Garment Texture Training Dataset},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/zyuzuguldu/lora-garment-textures}}
}