CGTrack: Cascade Gating Network with Hierarchical Feature Aggregation for UAV Tracking

TL;DR

This work tackles robust UAV tracking on edge devices by combining a lightweight hierarchical ViT backbone with a coarse-to-fine architecture. It introduces the Hierarchical Feature Cascade (HFC) to fuse multi-level features through feature reuse and residual gating, and the Lightweight Gated Center Head (LGCH) to extract fine-grained discriminative cues via Hadamard-product gating in an EG block. When integrated, CGTrack achieves state-of-the-art performance on UAV123, UAV123@10fps, and UAVTrack112 while maintaining high speed, validated by comprehensive ablations and qualitative results. The approach demonstrates that expanding network capacity through cascade gating and gating-based feature refinement can close the gap between efficient edge models and robust UAV tracking capabilities, with code to follow.

Abstract

Recent advancements in visual object tracking have markedly improved the capabilities of unmanned aerial vehicle (UAV) tracking, which is a critical component in real-world robotics applications. While the integration of hierarchical lightweight networks has become a prevalent strategy for enhancing efficiency in UAV tracking, it often results in a significant drop in network capacity, which further exacerbates challenges in UAV scenarios, such as frequent occlusions and extreme changes in viewing angles. To address these issues, we introduce a novel family of UAV trackers, termed CGTrack, which combines explicit and implicit techniques to expand network capacity within a coarse-to-fine framework. Specifically, we first introduce a Hierarchical Feature Cascade (HFC) module that leverages the spirit of feature reuse to increase network capacity by integrating the deep semantic cues with the rich spatial information, incurring minimal computational costs while enhancing feature representation. Based on this, we design a novel Lightweight Gated Center Head (LGCH) that utilizes gating mechanisms to decouple target-oriented coordinates from previously expanded features, which contain dense local discriminative information. Extensive experiments on three challenging UAV tracking benchmarks demonstrate that CGTrack achieves state-of-the-art performance while running fast. Code will be available at https://github.com/Nightwatch-Fox11/CGTrack.

Figures (8)

• Figure 1: Comparison of success rate and precision between CGTrack and other 13 state-of-the-art (SOTA) trackers on the authoritative UAV123 benchmark uav123. CGTrack achieves SOTA performance in both precision and success rate, surpassing the average performance of 13 trackers by 6.6% and 8.4% respectively. Best viewed in color for all figures in this paper.
• Figure 2: Comparison of the popular hierarchical feature fusion methods for UAV tracking. (a) Addition-Based Fusion: simply adds all the feature maps up. (b) Transformer-Based Fusion: employs Transformer layers or multi-head attention modules for feature fusion. (c) Cascade Gating Fusion: concatenates adjacent feature maps and performs gating subsequently in a cascade architecture.
• Figure 3: Overview of the proposed CGTrack, which comprises three main components: a lightweight hierarchical backbone, an HFC module, and a Lightweight Gated Center Head.
• Figure 4: Detailed architectures of LGCH. The left part illustrates the overall workflow of LGCH. The right one shows the structure of the EG block.
• Figure 5: Overall performance of CGTrack and prevailing SOTA trackers on UAV123 uav123 (the first column), UAV123@10fp uav123 (the second column), and UAVTrack112 uav112 (the third column) benchmarks. CGTrack achieves SOTA performance across all benchmarks.