WeexpectSwinTracktoserveasa solid baseline for Transformer tracking and facilitate future research.

Obviously, the self-attention module is permutation-invariance. Thus it can not "understand" the Before working with our tracker's encoder and decoder network, we need to extend the untied positional encoding to a multi-dimensional version. Together with relative positional bias, for an n-dimensional case, we have: α ij . From Tab. 1, we can observe The result shows our tracker is still competitive. Our tracker obtained the best performance on this benchmark.

Recently Transformer has been largely explored in tracking and shown state-of-the-art (SOT A) performance. However, existing efforts mainly focus on fusing and enhancing features generated by convolutional neural networks (CNNs). The potential of Transformer in representation learning remains under-explored.