Translation, Scale and Rotation: Cross-Modal Alignment Meets RGB-Infrared Vehicle Detection

Integrating multispectral data in object detection, especially visible and infrared images, has received great attention recent years. Since (RGB) (IR) images can provide complementary information to handle light variations, the paired are used many fields, such as pedestrian RGB-IR crowd counting salient detection. Compared with natural we find detection aerial suffers from cross-modal weakly misalignment problems, which manifested position, size angle deviations of same object. In this paper, mainly address challenge images. Specifically, firstly explain analyze cause problem. Then, propose a Translation-Scale-Rotation Alignment (TSRA) module problem by calibrating feature maps these two modalities. The predicts deviation between modality objects through an alignment process utilizes Modality-Selection (MS) strategy improve performance alignment. Finally, two-stream detector (TSFADet) based on TSRA is constructed for With comprehensive experiments public DroneVehicle datasets, verify that our method reduces effect achieve robust results.