Cityscape-Adverse is an extended benchmark designed to evaluate the robustness of semantic segmentation models under a variety of adverse environmental conditions. Using recent advancements in diffusion-based image editing, Cityscape-Adverse simulates realistic environmental changes such as weather variations, lighting shifts, and seasonal adjustments on the original Cityscapes dataset.

This project is a fork of the MMSegmentation repository from OpenMMLab. Cityscape-Adverse builds on MMSegmentation's capabilities, utilizing its modular framework to integrate and test models under challenging, synthesized conditions.

Note: The code and dataset will be made available soon. Stay tuned for updates!

This project builds upon MMSegmentation, an open-source semantic segmentation toolbox from OpenMMLab. We extend our gratitude to OpenMMLab for providing the foundational tools that enable Cityscape-Adverse.

For more updates, please stay tuned!