SparseTT: Visual Tracking with Sparse Transformers

TL;DR

SparseTT introduces a sparse attention mechanism and a double-head predictor to enhance visual tracking accuracy, outperforming state-of-the-art methods while reducing training time and maintaining real-time speed.

Contribution

The paper proposes a novel sparse attention mechanism and a double-head predictor for improved accuracy and efficiency in visual tracking with Transformers.

Findings

Outperforms state-of-the-art on multiple benchmarks

Runs at 40 FPS with reduced training time

Significantly improves tracking accuracy

Abstract

Transformers have been successfully applied to the visual tracking task and significantly promote tracking performance. The self-attention mechanism designed to model long-range dependencies is the key to the success of Transformers. However, self-attention lacks focusing on the most relevant information in the search regions, making it easy to be distracted by background. In this paper, we relieve this issue with a sparse attention mechanism by focusing the most relevant information in the search regions, which enables a much accurate tracking. Furthermore, we introduce a double-head predictor to boost the accuracy of foreground-background classification and regression of target bounding boxes, which further improve the tracking performance. Extensive experiments show that, without bells and whistles, our method significantly outperforms the state-of-the-art approaches on LaSOT, GOT-10k, TrackingNet, and UAV123, while running at 40 FPS. Notably, the training time of our method is reduced by 75% compared to that of TransT. The source code and models are available at https://github.com/fzh0917/SparseTT.