Abstract
RGBT tracking on edge devices is often constrained by limited computational power and data transmission bandwidth, necessitating the use of low-resolution inputs in real-world deployments. However, such resolution reduction typically leads to severe semantic loss and performance degradation. In this paper, we propose GAP-Track, an efficient framework that bridges the resolution gap by enabling high-precision tracking of low-resolution inputs. Specifically, we introduce a hierarchical knowledge distillation strategy that guides the low-resolution student by multi-level teacher supervision. To recover the missing semantic details, a generative reconstruction module trained by a random masking strategy is introduced to strengthen the student’s feature representation by forcing the backbone to recover fine-grained structural information from low resolution data. Furthermore, we adapt a polar-geometric sensitivity loss that transforms the bounding box regression into a polar coordinate system, effectively mitigating localization ambiguities in extremely low-resolution scenarios by providing higher sensitivity to center-point drift and shape deformations. Extensive experiments on the comprehensive LasHeR and RGBT234 datasets demonstrate that our method achieves state-of-the-art performance. Notably, GAP-Track delivers superior precision at 1/2 resolution and continues to significantly outperform the baseline even at 1/4 resolution, ensuring a robust balance between tracking accuracy and inference efficiency. The source code will be available at https://github.com/shijiahahaha/GAPTrack.