ParisStreetView-RandomMasks: Large-Scale Urban Image Inpainting Dataset with Random Irregular Masks

ParisStreetView-RandomMasks

ParisStreetView-RandomMasks is a large-scale urban image inpainting dataset designed for computer vision researchers, generative AI developers, and deep learning practitioners working on image restoration and scene completion tasks.

The dataset contains 22,601 street-view images along with synthetically generated irregular random masks and corresponding corrupted images for image inpainting research.

Dataset Contents

• Original urban street-view images
• Random irregular free-form masks
• Corrupted images for inpainting tasks
• Metadata annotations
• Training and validation split support

Dataset Structure

Dataset/
├── images/
├── masks/
├── corrupted/
├── train.txt
├── val.txt
└── annotations.csv

Applications

• Image Inpainting
• Diffusion Model Training
• Scene Completion
• Generative AI
• Computer Vision Research
• Image Restoration
• Deep Learning Benchmarking

Mask Generation

The masks were generated using irregular free-form random stroke simulation methods with varying thickness and region complexity to create realistic missing regions for supervised inpainting training.

Recommended Models

• LaMa
• Partial Convolution
• Stable Diffusion Inpainting
• EdgeConnect
• Context Encoder

Research Domains

• Computer Vision
• Generative AI
• Deep Learning
• Image Processing
• Scene Understanding

Acknowledgement

This work is based on the original Paris StreetView dataset. Additional preprocessing, random mask generation, and corrupted image creation were performed for image inpainting and diffusion model research purposes.

Please refer to the original dataset and authors for the base street-view imagery.

License

Please comply with the original dataset license and attribution requirements where applicable.

Additional resources related to deep learning, image processing, and computer vision research can be found at:

• Deep Learning Projects and Research Topics
• Image Processing and Computer Vision Projects