Learning Localization-Aware Target Confidence for Siamese Visual Tracking

Abstract

Siamese trackers based on 3D region proposal network (RPN) have shown remarkable success with deep Hough voting. However, using a single seed point feature as the cue for voting fails to produce high-quality 3D proposals. Additionally, the equal treatment of seed points in the voting process, regardless of their significance, exacerbates this limitation. To address these challenges, we propose a novel transformer-based voting scheme to generate better proposals. Specifically, a global-local transformer (GLT) module is devised to integrate object-and patch-aware geometric priors into seed point features, resulting in robust and accurate cues for offset learning of seed points. To train the GLT module, we introduce an importance prediction branch that learns the potential importance weights of seed points as a training constraint. Incorporating this transformerbased voting scheme into 3D RPN, a novel Siamese method dubbed GLT-T is developed for 3D single object tracking on point clouds. Moreover, we identify that the highest-scored proposal in the Siamese paradigm may not be the most accurate proposal, which limits tracking performance. Towards this concern, we approach the binary score prediction task as a ranking problem, and design a target-aware ranking loss and a localization-aware ranking loss to produce accurate ranking of proposals. With the ranking losses, we further present GLT-T++, an enhanced version of GLT-T. Extensive experiments on multiple benchmarks demonstrate that our GLT-T and GLT-T++ outperform state-ofthe-art methods in terms of tracking accuracy while maintaining a real-time inference speed. The source code will be made available at https://github.com/haooozi/GLT-T.

Multi-Correlation Siamese Transformer Network With Dense Connection for 3D Single Object Tracking

“…Results on KITTI. We compare our algorithm with ten stateof-the-art algorithms [2]- [7], [11], [12], [43], [44] on the KITTI dataset [17]. Besides M 2 -Track [7] and CXTrack [43] which adopt a motion-centric (MC) paradigm and a full framebased similarity (FFS) paradigm, respectively, our algorithm and all other compared algorithms use a region-based similarity paradigm.…”

GLT-T: Global-Local Transformer Voting for 3D Single Object Tracking in Point Clouds

“…Transformer is originally proposed in the area of natural language processing (Vaswani et al 2017), showing an excellent ability in modeling long-range dependency. It becomes popular in computer vision recently and has been widely used for image classification (Liu et al 2021;Xu et al 2021;Zhang et al 2022a,b), object detection (Carion et al 2020;Wang et al 2021aWang et al , 2022, pose estimation (Xu et al 2022), and object tracking (Lan et al 2022;Nie et al 2022a). The success motivates researchers to extend it to 3D point cloud tasks.…”

DFSA-DAN: dynamic fusion of statistical metric and adversarial learning for domain adaptation network based intelligent fault diagnosis

“…Among the array of deep learning approaches stand out as exemplary, demonstrating exceptional performance in fault diagnosis. Some typical deep learning models, for example, convolutional neural network (CNN) [6][7][8], deep belief network [9] and deep auto-encoder [10], already employed in fault diagnosis effectively. However, the effectiveness of deep learning applications is contingent upon a fundamental assumption, the learning and testing data adhere to a same distribution, originating from the same operating conditions [11].…”