Główne pojęcia
Proposing C2Former to address modality miscalibration and fusion imprecision in RGB-IR object detection.
Streszczenie
Object detection using RGB and IR images is crucial for around-the-clock applications. Existing methods face challenges of modality miscalibration and fusion imprecision. C2Former addresses these issues with an Inter-modality Cross-Attention module and Adaptive Feature Sampling. Experimental results show improved detection accuracy on DroneVehicle and KAIST datasets.
Statystyki
Extensive experiments conducted on DroneVehicle and KAIST RGB-IR datasets.
C2Former achieves 74.2% mAP, outperforming other multispectral detection methods.
Computational cost reduced from 139.8G FLOPs to 100.9G FLOPs with the addition of AFS module.
Cytaty
"In C2Former, we design an Inter-modality Cross-Attention (ICA) module to obtain the calibrated and complementary features by learning the cross-attention relationship between the RGB and IR modality."
"With extensive experiments on the DroneVehicle and KAIST RGB-IR datasets, we verify that our method can fully utilize the RGB-IR complementary information."