Low-Complexity Multi-Type Tree Partitioning For Versatile Video Coding Based On Machine Learning

One of the main issues of current detection-based tracking methods is object annihilation, which is the object lost in one or more frames caused by errors of the object detector, especially in adverse weather or occluded scenes. in this work, the matter waves network (MWNET) based on quantum theory is proposed to improve tracking robustness. The object detection results are encoded into the matter waves by quantum measurement to utilize the features of the complex value. The object is regenerated from annihilation by quantum evolution to repair the detection errors. The network can improve the information flow between front and back frames and the continuity of tracking trajectory. Finally, on the MOT20 test set, 77.3 MOTA and 76.2 IDF1 are implemented by MWNET. Our work proves the effectiveness of quantum evolution in reconstructing information, and the model can realize continuous tracking of multiple objects in frequent occlusion scenes.