Part‐MOT: A multi‐object tracking method with instance part‐based embedding

Abstract

Part-MOT, a one-stage anchor-free architecture which unifies the object identification representation and detection in one task for visual object tracking is presented. For object representation, a position relevant feature is obtained using the center-ness information, which takes advantage of the anchor-free ideal to encode the feature map as the instanceaware embedding. To adapt to the object's movement, the clustering-based method to get the global instance feature is introduced. This enables this approach more robust to make better tracking decisions. Part-MOT achieves the state-of-the-art performance on public datasets, with especially strong results for object deformation and movement changes.This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Semantic‐guided fusion for multiple object tracking and RGB‐T tracking

“…The illustration of the RGB-T tracker with SGF module, where the semantic-guided fusion block is the same as Figure 3, and the transformation function is shown in Equation (10).…”

“…Following [9,10], we created a large training dataset by combining training images from the MOT16 and MOT17 datasets. We use the similar architecture as the FairMOT with the SGF module used to encode features.…”

“…However, the performance was lower than the two‐step models, because the Re‐ID features usually require more low‐level features to discriminate different instances of the same class, and the detection features must be similar for different instances. Improved methods have been proposed with detailed feature fusion designs with a more complicated encoder structure, such as FairMOT [9], Part‐MOT [10], MANet [11] etc. The third paradigm is the attention‐based tracking, as shown in Figure 1c.…”

One-Shot Multiple Object Tracking in UAV Videos Using Task-Specific Fine-Grained Features

“…Meng et al [37] presented a spatio-temporal attention for updating the weights of identity embedding at each moment. Liu et al [38] designed a deformable convolution-based region transformation module to reduce the focus on irrelevant regions in the ReID branch. Yan et al [39] employed FPN to aggregate multi-level features to enrich the information of targets.…”