MVDet: multi-view multi-class object detection without ground plane assumption

Abstract Although many state-of-the-art methods of object detection in a single image have achieved great success the last few years, they still suffer from false positives crowd scenes real-world applications like automatic checkout. In order to address limitations single-view complex scenes, we propose MVDet, an end-to-end learnable approach that can detect and re-identify multi-class objects multiple images captured by cameras (multi-view). Our is based on premise incorrect results specific view be eliminated using precise cues other views, given availability multi-view images. Unlike most existing algorithms, which assume belong class ground plane, our classify without such assumptions thus more practical. To objects, integrated architecture for region proposal, re-identification, classification. Additionally, utilize epipolar geometry constraint devise novel re-identification algorithm does not require about plane assumption. model demonstrates competitive performance compared several baselines challenging MessyTable dataset.