SwinTrack: A Simple and Strong Baseline for Transformer Tracking

Transformer has recently demonstrated clear potential in improving visual tracking algorithms. Nevertheless, existing transformer-based trackers mostly use Transformer to fuse and enhance the features generated by convolutional neural networks (CNNs). By contrast, in this paper, we propose a fully attentional-based Transformer tracking algorithm, Swin-Transformer Tracker (SwinTrack). SwinTrack uses Transformer for both feature extraction and feature fusion, allowing full interactions between the target object and the search region for tracking. To further improve performance, we investigate comprehensively different strategies for feature fusion, position encoding, and training loss. All these efforts make SwinTrack a simple yet solid baseline. In our thorough experiments, SwinTrack sets a new record with 0.717 SUC on LaSOT, surpassing STARK by 4.6\% while still running at 45 FPS. Besides, it achieves state-of-the-art performances with 0.483 SUC, 0.832 SUC and 0.694 AO on other challenging LaSOT$_{ext}$, TrackingNet, and GOT-10k. Our implementation and trained models are available at https://github.com/LitingLin/SwinTrack.