kromic/sd-crosswalk-augmentation

Model Description

This model is a fine-tuned Stable Diffusion model to generate realistic pedestrian-perspective images of crosswalks. It was fine-tuned on a dataset of 150 first-person view (FPV) images, primarily captured in sunny conditions, to enable controlled text-to-image generation for data augmentation in crosswalk segmentation tasks.

• Base model: Stable Diffusion v1.4
• Fine-tuning method: Text-to-image fine-tuning using custom FPV crosswalk dataset
• Components:
  • unet — fine-tuned U-Net weights
  • vae — fine-tuned VAE weights
• Intended use: Synthetic data generation for semantic segmentation augmentation

---

Use Cases

• Data augmentation for crosswalk segmentation models
• Generating diverse weather and lighting scenarios (e.g., fog, rain, snow, night) from text prompts
• Research on assistive navigation systems for visually impaired pedestrians
• Benchmarking model generalization across diverse environments

---

How to Use

You can generate images with the provided Python inference script:

# Clone the repository
git clone https://huggingface.co/kromic/sd-crosswalk-augmentation
cd sd-crosswalk-augmentation

# Install dependencies
pip install diffusers transformers torch

# Run inference
python generate.py

# Customize your prompt
prompt = "a crosswalk image"