ReDAM and Unified-weather-edit Align Multi-Sensor Weather Simulations for AVs
A new arXiv preprint (2607.25612) introduces two methods to align multi-sensor weather simulations for autonomous vehicles. The Reference Dataset Alignment Method (ReDAM) aligns fog intensity, while Unified-weather-edit (inspired by Weather-edit) aligns particle positioning in rain and snow. Validation used statistical and geometrical tests. Non-aligned simulations make 3D detection models overly optimistic compared to aligned versions. Fine-tuning on aligned multi-sensor data improves robustness for 3D object detection.
Key facts
- arXiv preprint 2607.25612 proposes ReDAM and Unified-weather-edit for weather simulation alignment.
- ReDAM aligns fog intensity across sensors.
- Unified-weather-edit aligns particle positioning in rain and snow.
- Validation uses statistical and geometrical tests.
- Non-aligned simulations lead to overly optimistic 3D detection models.
- Aligned simulations improve robustness for 3D object detection via fine-tuning.
Entities
Institutions
- arXiv