Weak supervision crowd counting system and method guided by multi-scale cross-modal similarity

The invention discloses a multi-scale cross-modal similarity guided weak supervision crowd counting system and method, and relates to the technical field of computer vision, and the method comprises the following steps: obtaining a visible light image and a thermal infrared image of a to-be-counted scene; inputting the visible light image and the thermal infrared image into a multi-scale cross-modal similarity guided weak supervision crowd counting system to obtain a crowd number estimation result; the multi-scale cross-modal similarity guided weak supervision crowd counting system comprises a modal specific feature extraction module, a multi-scale context feature fusion module, a crowd counting prediction module and a parameter optimization module which are connected in sequence. The cross-modal crowd counting method can effectively solve the problems that the counting error is large due to the fact that the appearance scale of the crowd changes drastically and a large amount of fine annotation information i