Abstract
Perception tasks for autonomous vehicles need to work satisfactorily in adverse weather conditions. Due to lack of real-world weather datasets, weather simulations are a promising alternative. To ensure simulations closely mirror real-world weather data, it's crucial that they represent the same weather characteristics, including severity and particle positioning, across different sensors.
To achieve this, we propose the Reference Dataset Alignment Method (ReDAM) for weather intensity alignment in fog and Unified-weather-edit (inspired by Weather-edit) for particle positioning alignment in rain and snow. We validate both alignment methods using statistical and geometrical tests, respectively.
We find that 3D detection models for non-aligned versions tend to be overly optimistic as compared to aligned versions. We also show the aligned multi-sensor simulation's effectiveness for achieving robustness for the 3D object detection task by fine-tuning existing sensor fusion models on it.
Blogger's Review: The proposed alignment methods offer a fresh solution for enhancing perception in autonomous driving under adverse weather conditions, significantly improving the realism of simulation data. Their innovative and practical approach sets a solid foundation for future research in weather simulations.