WILD SAM: A Simulated-and-Real Data Augmentation for Autonomous Driving Perception under Challenging Weather

Name: WILD SAM: A Simulated-and-Real Data Augmentation for Autonomous Driving Perception under Challenging Weather
Uploaded: 2026-10-13T12:02:50.598580+00:00
Duration: 316 s
Description: The WILD SAM framework combines physics-based simulation and pseudo-label denoising to improve 3D object detection performance under adverse weather like rain and snow. It trains models using real harsh weather data, augmented by simulated frames and cleaned pseudo-labels, achieving up to a 13% increase in accuracy over baseline methods.

The WILD SAM framework combines physics-based simulation and pseudo-label denoising to improve 3D object detection performance under adverse weather like rain and snow. It trains models using real harsh weather data, augmented by simulated frames and cleaned pseudo-labels, achieving up to a 13% increase in accuracy over baseline methods.

Open the full episode