Most calibration methods operate at the logit level, implicitly assuming that miscalibration can be corrected without changing the underlying representation. We challenge this assumption and propose \textbf{Calibration Attention (CalAttn)}, a \emph{representation-aware} calibration module for vision transformers that couples instance-wise temperature scaling to transformer token geometry under a proper scoring objective. CalAttn predicts a sample-specific temperature from the \texttt{[CLS]} token and backpropagates calibration gradients into the backbone, thereby reshaping the uncertainty structure of the representation rather than post-hoc adjusting confidence. This yields \emph{token-conditioned uncertainty modulation} with negligible overhead (\(<0.1\%\) additional parameters). Across multiple datasets with ViT/DeiT/Swin backbones, CalAttn consistently improves calibration while preserving accuracy, achieving relative ECE reductions of \(3.7\%\) to \(77.7\%\) over strong baselines across diverse training objectives. Our results indicate that treating calibration as a representation-level problem is a practical and effective direction for trustworthy uncertainty estimation in transformers. Code: [https://github.com/EagleAdelaide/CalibrationAttention-CalAttn-](https://github.com/EagleAdelaide/CalibrationAttention-CalAttn-)

翻译：大多数校准方法在逻辑层面进行操作，隐式地假设可以在不改变底层表征的情况下校正失准。我们挑战这一假设，并提出**校准注意力（CalAttn）**，一种用于视觉Transformer的**表征感知**校准模块，它在适当的评分目标下将实例级温度缩放与Transformer令牌几何结构耦合。CalAttn从`[CLS]`令牌预测样本特定的温度，并将校准梯度反向传播到骨干网络中，从而重塑表征的不确定性结构，而非事后调整置信度。这实现了**令牌条件化的不确定性调制**，且开销可忽略（额外参数<0.1%）。在采用ViT/DeiT/Swin骨干网络的多个数据集上，CalAttn在保持准确性的同时持续改善校准效果，相较于强基线方法，在不同训练目标下实现了3.7%至77.7%的相对ECE降低。我们的结果表明，将校准视为表征层面的问题是实现Transformer可信不确定性估计的一个实用且有效的方向。代码：[https://github.com/EagleAdelaide/CalibrationAttention-CalAttn-](https://github.com/EagleAdelaide/CalibrationAttention-CalAttn-)