自适应深度轻量级 RGB-T 跟踪与整体令牌路由
Adaptive Depth Lightweight RGB-T Tracking with Holistic Token Routing
自适应早退策略和整体令牌引导的跨模态交互模块可被复用于其他多模态跟踪或轻量化视觉任务，在保持精度的同时显著降低计算量和延迟。
原文依据 we propose Adaptive Early-Exit (AEE): we augment the backbone with anytime heads and pair them with a confidence-calibrated early-exit policy that halts inference at the earliest reliable layer
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论RGB-T 跟踪的吸引力在于其在 RGB 失效时（如夜间、眩光、雾天和部分遮挡）的鲁棒性。尽管精度显著提升，但近期架构强调深度融合和大参数量，导致浮点运算量和带宽增加。这种计算负担限制了实时性能，并限制了在高性能 GPU 之外的扩展性。为了平衡精度和效率，我们提出了自适应早退（AEE）：我们通过随时预测头增强骨干网络，并将其与置信度校准的早退策略配对，该策略在最早可靠层停止推理，跳过冗余计算。对于跨模态交互，我们设计了一个整体令牌引导的交互（HTGI）模块，其中每个模态被压缩为一组紧凑的整体状态令牌，并注入到另一模态的建模流中，无需逐层对齐，从而以极低成本实现有针对性的信息交换。在 RGB-T 基准上，轻量级跟踪器在保持有竞争力的精度的同时大幅降低了延迟；在 LasHeR 上，它实现了 70.2% 的精度和 56.3% 的成功率，在 GPU 上以 148.3 FPS 运行，在 CPU 上以 50.2 FPS 运行，在边缘设备上以 28.7 FPS 运行。
英文原文摘要 1189 字符
The appeal of RGB-T tracking lies in its resilience when RGB fails under night scenes, glare, fog, and partial occlusion. Despite notable accuracy gains, recent architectures emphasize deep fusion and large parameter counts, driving up FLOPs and bandwidth. This computational burden constrains real-time performance and limits scalability beyond high-end GPUs. To balance accuracy and efficiency, we propose Adaptive Early-Exit (AEE): we augment the backbone with anytime heads and pair them with a confidence-calibrated early-exit policy that halts inference at the earliest reliable layer, skipping redundant computation. For cross-modal interaction, we design a Holistic-Token-Guided Interaction (HTGI) module, where each modality is compressed into a compact set of holistic state tokens and injected into the other modality's modeling stream without layer-wise alignment, enabling targeted information exchange at extremely low cost. On RGB-T benchmarks, the lightweight tracker substantially reduces latency while maintaining competitive accuracy; on LasHeR, it achieves 70.2% precision and 56.3% success, running at 148.3 FPS on GPU, 50.2 FPS on CPU, and 28.7 FPS on an edge device.