Extended ablation study results on the nuScenes validation set. This table compares the performance of LGMMfusion variants, including the impact of Image BEV (I-BEV), BEV Query (BEV-Q), attention structure modules (MHMS-SA and MHA-CA), and attention parameters (number of heads H and scales S). I-BEV variations involve the use of BatchNorm (BN) and ReLU activations.

Extended ablation study results on the nuScenes validation set. This table compares the performance of LGMMfusion variants, including the impact of Image BEV (I-BEV), BEV Query (BEV-Q), attention structure modules (MHMS-SA and MHA-CA), and attention parameters (number of heads H<...

Full description

Saved in:

Bibliographic Details
Main Author:	Haixing Cheng (2240002) (author)
Other Authors:	Chengyong Liu (9976282) (author), Wenzhe Gu (5802653) (author), Yuyi Wu (21807702) (author), Mengye Zhao (22184576) (author), Wentao Liu (1447222) (author), Naibang Wang (22184579) (author)
Published:	2025
Subjects:	Neuroscience Cancer Science Policy Space Science Biological Sciences not elsewhere classified Information Systems not elsewhere classified nuscenes validation set experimental results show resolution image features provide enhanced features better align image approaches often fail head adaptive cross sparse lidar points xlink "> multi prior depth information lgmmfusion achieves 71 image bev features modal fusion framework lidar bev features depth information lidar features often lead modal sensors framework leverages better match lidar bev head multi novel lidar detection head spatial positions small objects scale self point clouds perception systems imprecise detections guided multi fully exploit feature fusion eye view critical role called lgmmfusion autonomous driving attention mechanism also improving
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items