Alapfogalmak
Proposing a novel UDA method to refine dynamic and small objects for nighttime semantic segmentation.
Kivonat
Nighttime semantic segmentation is crucial for applications like autonomous driving, facing challenges due to illumination conditions. Existing UDA methods struggle with dynamic and small objects. The proposed method refines these objects at label and feature levels, outperforming prior arts in nighttime segmentation.
Statisztikák
Extensive experiments show mean intersection over union (mIoU) scores of 60.9% for 'pole', 86.8% for 'car', and 25.2% for 'bus'.
Proposed method achieves a significant improvement of 2.9%, 3.0%, and 20.2% absolute performance over the state-of-the-art method.
Idézetek
"We propose a novel UDA framework focusing on dynamic and small object refinements."
"Our approach achieved remarkable mIoU scores in identifying dynamic and small objects."