Weather-nuScenes: A Synthetic AI-Generated Weather-Augmented Derivative of the nuScenes Dataset for Autonomous Driving

Weather-nuScenes Weather-nuScenes is a large-scale synthetic data augmentation framework and dataset developed by researchers at the Information Processing and Telecommunications Center of the Universidad Politécnica de Madrid and the Technische Universität Berlin. It is derived from the original nuScenes dataset. By leveraging generative multimodal models via the Gemini API using the Nano Banana configuration, this project systematically transforms synchronized 6-camera vehicle images into synthetic weather and environmental variations. The objective is to support the benchmarking of autonomous driving perception models under extreme out-of-distribution visual conditions. A sample of the images can be visualized at: https://ging.github.io/Weather-nuScenes/ Target Weather Conditions The pipeline transforms standard clear-weather multi-camera frames into seven distinct environmental variations: Fog: High-density volumetric occlusion and reduced contrast. Night: Low-light degradation with active vehicle and street lighting signatures. Night Rain: Wet surface reflections, glare, and low-light noise profiles. Night Snow: Low-light accumulation, ground coverage, and falling flakes. Rain: Wet asphalt specular reflections, spray patterns, and lens droplets. Sandstorm: Heavy warm-toned particulate haze and severely restricted visibility. Snow: High-albedo ground coverage and winter scene textures. Repository and File Inventory peticion_nano_banana.py Description: Core execution script interfacing with the Gemini API to handle pipeline generation, sequential checkpointing, and fault isolation. peticion_nano_banana_config.example.py Description: Blueprint configuration file detailing API parameters, prompt strings, and target file mapping. README_peticion_nano_banana.md Description: Specialized technical documentation providing onboarding steps, retry policies, and syntax definitions for the generator script. sample_and_viewer_example.zip Description: Public Access: A lightweight verification package containing a miniature sample structure and the visualization script. sample_and_viewer.zip Description: Restricted Access: The full high-fidelity synthetic dataset containing all 6 surround-view camera variations alongside integration utilities. Data Generation Workflow The transformation engine operates by pulling raw imagery from the 6-camera arrays located in: ImagenesOriginales/nuScenes_val_6_cams/ It then runs asynchronous transformation queries through the model and commits the generated outputs to structured folders under: Resultados/ejecucionNanoBanana_22062026/ For full setup, optimization, and recovery flags, please consult the standalone documentation file: README_peticion_nano_banana.md Terms of Use and Citations This dataset is a derivative work of the original nuScenes dataset. In accordance with the nuScenes Terms of Use, the dataset contents are distributed strictly under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license: CC BY-NC-SA 4.0 The accompanying code tools are available under open Creative Commons Attribution 4.0 International terms: CC BY 4.0 If you leverage this dataset or generation engine in an academic setting, please cite both this repository record and the underlying foundational work. This Dataset IEEE style: [1] J. D. Molero García-Morato, J. Conde Díaz, G. Martínez Ruiz de Arcaute, C. Arriaga, P. Reviriego, J. Yuan, D.-M. Nguyen, D. Le Phuoc, and A. Le Tuan, "Weather-nuScenes: A Synthetic AI-Generated Weather-Augmented Derivative of the nuScenes Dataset for Autonomous Driving," *Zenodo*, Jul. 02, 2026. DOI: 10.5281/zenodo.21135950. BibTeX: @dataset{weather_nuscenes_2026, author = {Molero García-Morato, J. D. and Conde Díaz, J. and Martínez Ruiz de Arcaute, G. and Arriaga, C. and Reviriego, P. and Yuan, Jicheng and Nguyen, Duc-Manh and Le Phuoc, Danh and Le Tuan, Anh}, title = {Weather-nuScenes: A Synthetic AI-Generated Weather-Augmented Derivative of the nuScenes Dataset for Autonomous Driving}, publisher = {Zenodo}, year = {2026}, month = jul, doi = {10.5281/zenodo.21135950} } Original nuScenes Framework @article{ title={nuScenes: A multimodal dataset for autonomous driving}, author={Holger Caesar and Varun Bankiti and Alex H. Lang and Sourabh Vora and Venice Erin Liong and Qiang Xu and Anush Krishnan and Yu Pan and Giancarlo Baldan and Oscar Beijbom}, journal={arXiv preprint arXiv:1903.11027}, year={2019} } Acknowledgements Access to the Nanobanana model was provided by the Google Cloud Research Credits program under the Gemini Academic Program. This work was supported by the Agencia Estatal de Investigación (AEI) (doi:10.13039/501100011033) under Grants FUN4DATE (PID2022-136684OB-C22) and SMARTY (PCI2024-153434) and by the European Commission through the Chips Act Joint Undertaking project SMARTY (Grant 101140087).