Abstract:Over the last decade, Siamese network architectures have emerged as dominating tracking paradigms, which have led to significant progress. These architectures are made up of a backbone network and a head network. The backbone network comprises two identical feature extraction sub-branches, one for the target template and one for the search candidate. The head network takes both the template and candidate features as inputs and produces a local similarity score for the target object in each location of the search candidate. Despite promising results that have been attained in visual tracking, challenges persist in developing efficient and lightweight models due to the inherent complexity of the task. Specifically, manually designed tracking models that rely heavily on the knowledge and experience of relevant experts are lacking. In addition, the existing tracking approaches achieve excellent performance at the cost of large numbers of parameters and vast amounts of computations. A novel Siamese tracking approach called TrackNAS based on neural architecture search is proposed to reduce the complexity of the neural architecture applied in visual tracking. First, according to the principle of the Siamese network, backbone and head network search spaces are constructed, constituting the search space for the network architecture. Next, under the given resource constraints, the network architecture that meets the tracking performance requirements is obtained by optimizing a hybrid search strategy that combines distributed and joint approaches. Then, an evolutionary method is used to lighten the network architecture obtained from the search phase to facilitate deployment to devices with resource constraints (FLOPs). Finally, to verify the performance of TrackNAS, comparison and ablation experiments are conducted using several large-scale visual tracking benchmark datasets, such as OTB100, VOT2018, UAV123, LaSOT, and GOT-10k. The results indicate that the proposed TrackNAS achieves competitive performance in terms of accuracy and robustness, and the number of network parameters and computation volume are far smaller than those of other advanced Siamese trackers, meeting the requirements for lightweight deployment to resource-constrained devices.

Fast Visual Tracking with Lightweight Siamese Network and Template-Guided Learning

Improving Siamese Based Trackers with Light or No Training through Multiple Templates and Temporal Network

Real-time object tracking in the wild with Siamese network

Efficient and Lightweight Visual Tracking with Differentiable Neural Architecture Search

Masked and Dynamic Siamese Network for Robust Visual Tracking.

Light Siamese Network for Long-Term Onboard Aerial Tracking

Distractor-Aware Visual Tracking by Online Siamese Network

Learning Dynamic Siamese Network For Visual Object Tracking

Hybrid-Mode Tracker with Online SA-LSTM Updater

Dynamic template updating Siamese network based on status feedback with quality evaluation for visual object tracking

Siamese Residual Network for Efficient Visual Tracking

Visual Object Tracking via an Improved Lightweight Siamese Network

Efficient object tracking algorithm based on lightweight Siamese networks

Online visual tracking via background-aware Siamese networks

Faster and Simpler Siamese Network for Single Object Tracking

Distillation, Ensemble and Selection for building a Better and Faster Siamese based Tracker

Siamese Network with Contrastive Learning and Adaptive Template Updating for Object Tracking.

A Location-Aware Siamese Network for High-Speed Visual Tracking

Visual Tracking With Siamese Network Based on Fast Attention Network

Extremely Tiny Siamese Networks with Multi-level Fusions for Visual Object Tracking.

Single Target Tracking Algorithm for Lightweight Siamese Networks Based on Global Attention