Local Feature Enhancement based One-Stream Transformer Tracking

In target tracking, most existing one-stream Transformer methods focus on the extraction of global features and lack attention to local features at the detail level. To address this problem, this paper proposes a one-stream Transformer tracking method based on local feature enhancement. The method further effectively enhances the utilization of detail features by directly extracting local features from the search region and using these local windows to enhance the search region. This method is able to focus on the detail level of the target while preserving the contextual information, thus achieving an effective balance between global perception and local characterization capabilities. The experimental results show that this method improves the tracker performance over the traditional global enhancement model on benchmark datasets such as TrackingNet, GOT-10K and TNL2K.