让视觉Transformer的注意力机制可解释，且不影响精度。
Interpretable Vision Transformers in Monocular Depth Estimation via SVDA
- 引入基于SVD的注意力机制，分离方向对齐与频谱调制。
- 在KITTI和NYU-v2上保持精度，仅增加少量计算开销。
- 提供6个可量化指标，揭示注意力的内在规律。
单目深度估计是计算机视觉中的核心问题，广泛应用于机器人、AR和自动驾驶领域，但现代Transformer架构中的自注意力机制仍不透明。本文将受SVD启发的注意力（SVDA）引入密集预测Transformer（DPT），首次为密集预测任务提供了谱结构化的注意力形式。SVDA通过在归一化查询-键交互中嵌入可学习对角矩阵，将方向对齐与频谱调制解耦，使注意力图具有内在可解释性，而非事后近似。在KITTI和NYU-v2数据集上的实验表明，SVDA在保持或略微提升预测精度的同时，仅带来微小的计算开销。更重要的是，SVDA解锁了六种谱指标：熵、秩、稀疏性、对齐度、选择性和鲁棒性，揭示了训练过程中注意力组织的一致跨数据集和深度依赖模式，这些洞见在标准Transformer中无法获取。通过将注意力从黑箱机制转变为可量化的描述符，SVDA重新定义了单目深度估计中的可解释性，并为透明密集预测模型开辟了原则性路径。
原文摘要 · Abstract (English)
Monocular depth estimation is a central problem in computer vision with applications in robotics, AR, and autonomous driving, yet the self-attention mechanisms that drive modern Transformer architectures remain opaque. We introduce SVD-Inspired Attention (SVDA) into the Dense Prediction Transformer (DPT), providing the first spectrally structured formulation of attention for dense prediction tasks. SVDA decouples directional alignment from spectral modulation by embedding a learnable diagonal matrix into normalized query-key interactions, enabling attention maps that are intrinsically interpretable rather than post-hoc approximations. Experiments on KITTI and NYU-v2 show that SVDA preserves or slightly improves predictive accuracy while adding only minor computational overhead. More importantly, SVDA unlocks six spectral indicators that quantify entropy, rank, sparsity, alignment, selectivity, and robustness. These reveal consistent cross-dataset and depth-wise patterns in how attention organizes during training, insights that remain inaccessible in standard Transformers. By shifting the role of attention from opaque mechanism to quantifiable descriptor, SVDA redefines interpretability in monocular depth estimation and opens a principled avenue toward transparent dense prediction models.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。