Reproduce “WeatherCity: Urban Scene Reconstruction with Controllable Multi-Weather Transformation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Wenhua Wu, Huai Guan, Zhe Liu, Hesheng Wang

WeatherCity creates multi-weather renderings of a reconstructed dynamic street scene while keeping its underlying Gaussian representation shared across weather conditions. Text-guided image editing supplies supervision, and weather-specific decoders change appearance; rain, snow, and fog are added through controllable simulation. On Waymo’s rainy-weather comparison, it outscored Qwen-Image on CLIP-S, CLIP-DS, and Sem-CS; this supports image and semantic similarity, not driving performance.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.