Beyond Traditional Single Object Tracking: A Survey

TL;DR

This survey analyzes non-traditional deep-learning techniques applied to single object tracking, framing a taxonomy that includes Sequence Models, Generative Models, Self-supervised/Unsupervised Learning, Meta-Learning, Continual Learning, and Domain Adaptation. It contrasts autoregressive versus memory-based sequence models, and surveys generative approaches (GANs, VAEs, diffusion, masked modelling) as tools to enhance robustness and generalization. It also synthesizes a broad experimental analysis across datasets such as GOT-10k, LaSOT, TrackingNet, and VOT, using metrics AO, $SR_{0.5}$, $SR_{0.75}$, $AUC$, $P$, $P_{Norm}$, and $EAO$. The discussion highlights transformer-dominated trends, strong benefits from temporal coherence and pretraining, and identifies promising future directions including joint autoregressive-temporal modelling and diffusion-based data augmentation for SOT.

Abstract

Single object tracking is a vital task of many applications in critical fields. However, it is still considered one of the most challenging vision tasks. In recent years, computer vision, especially object tracking, witnessed the introduction or adoption of many novel techniques, setting new fronts for performance. In this survey, we visit some of the cutting-edge techniques in vision, such as Sequence Models, Generative Models, Self-supervised Learning, Unsupervised Learning, Reinforcement Learning, Meta-Learning, Continual Learning, and Domain Adaptation, focusing on their application in single object tracking. We propose a novel categorization of single object tracking methods based on novel techniques and trends. Also, we conduct a comparative analysis of the performance reported by the methods presented on popular tracking benchmarks. Moreover, we analyze the pros and cons of the presented approaches and present a guide for non-traditional techniques in single object tracking. Finally, we suggest potential avenues for future research in single-object tracking.

Figures (8)

• Figure 1: A timeline of the most important trackers per each of the six broad categories of this survey.
• Figure 2: Literature Survey Tree
• Figure 3: The autoregressive family of trackers. The template size can be different from one tracker to another. Some trackers use the full resolution of the template while others use a smaller size that focuses on the object of interest.
• Figure 4: A general diagram of memory-based sequence models.
• Figure 5: A general pertaining scheme using a masked autoencoder in single object tracking.