Dynamic token sampling for efficient unmanned aerial vehicles transformer tracking

The existing Transformer based unmanned aerial vehicles tracking methods suffer from issues such as token redundancy and lack of information. In order to solve the above problems, we propose a novel dynamic token sampling for an efficient unmanned aerial vehicle transformer tracking framework. Unlike previous transformer-based tracking methods, we avoids the need for complex head networks like classification and regression. We design encoders that consist of three key components: Dynamic Position Embedding, Dynamic Token Sampler, and Convolutional Feed-Forward Network. This module enhances visual representation by scoring and dynamically sampling tokens, allowing for a flexible token count that adapts to target changes within each frame. We utilize a simple image-sequence contrastive loss as the loss function. Our approach not only simplifies the tracking framework but also achieves state-of-the-art performance at real-time running speeds across seven challenging datasets.