Cost Volume.
The cost volume technique has been success-
fully applied in depth estimation [11, 55, 18] and optical
flow estimation [35, 10, 53] for associating pixels between
two frames. This motivates us to extend cost volume to a
multi-object tracker, which will be demonstrated to be ef-
fective in learning re-ID embeddings and inferring tracking
offsets in this paper. Our approach may inspire future works
using cost volume in tracking or re-identification.
Cost Volume结构
这项技术已经在深度估计[11,55,18]和光流估计[35,10,53],用于估计两帧之间的关联像素。这促使我们将cost volume结构扩展到多目标跟踪器,这将在本文中被证明对于学习reid的嵌入和推断其跟踪偏移量是有效的。我们的方法可能会启发到未来使用这项技术进行跟踪或者reid的工作.
4. TraDeS Tracker
our idea
以前的大多数联合检测和跟踪方法都是独立检测,而不需要跟踪的显式输入。相比之下,我们的目标是将跟踪线索整合到端到端检测中,从而提高对艰难场景的检测,反过来也有利于跟踪。为此,我们提出了一个基于成本量的关联(CVA:§4.1)模块来学习reid嵌入和推断对象运动,