[ML-news] [Software] [Dataset] Release of Snowy Scenes - A Multimodal Multitask Dataset Toward Snow-tonomous Vehicles

[Eren Erdal Aksoy]

Dear  Colleagues,

Autonomous vehicles perform impressively well in clear weather, but their performance in snowy conditions remains a critical challenge. Snowflakes confuse LiDAR, road markings disappear, and numerous false-positive ghost objects emerge.

That's why we're excited to release "Snowy Scenes: A Multimodal Multitask Dataset Toward Snow-tonomous Vehicles" (published in IEEE RA-L, 2026). 

The dataset was collected entirely under real-world snowy conditions in Finland and provides 22K multimodal data to advance robust perception in winter driving.

Snowy Scenes has comprehensive human annotations for more than 5K sequential LiDAR scans, including 221K precise bounding boxes for object detection, per-point semantic segmentation, and the labeling of falling snow particles.

Some highlights:
❄️ Real snowy urban and highway driving scenarios
📡 128-beam LiDAR, RGB, and thermal cameras together with GNSS/IMU
🚘 3D bounding box as well as point-wise annotations, even for falling snowflakes 
🔀 Joint learning of de-snowing, semantic segmentation, and object detection tasks

Paper: https://ieeexplore.ieee.org/abstract/document/11623683

Code: https://github.com/snowyscenes/dataset

Video: https://www.youtube.com/playlist?list=PLeeRDpXIlOqo

For any questions, feedback, or suggestions, please feel free to reach out.

Best

Eren

 ---

Dr. rer. nat. Eren Erdal Aksoy, Associate Professor

Department of Computer Science, LTH, Lund University

Ole Römers väg 3, Box 118, 22100 Lund, Sweden 

Phone: +46 70 610 90 75

E-Mail: eren....@cs.lth.se

Web: https://aksoyeren.github.io/