提出量化图像形状与纹理比例的方法，提升卷积网络在草图等形状主导数据上的表现。
Quantifying and Inducing Shape Bias in CNNs via Max-Pool Dilation
- 通过亮度通道与L0平滑版本的SSIM值衡量数据集形状纹理平衡度。
- 仅调整最大池化层膨胀率，在低数据场景下显著提升草图类数据准确率。
- 无需重训练卷积权重，适合资源受限或小样本场景使用。
卷积神经网络（CNN）普遍存在纹理偏好，即更关注局部模式而非整体形状，这源于其卷积结构。虽然对纹理丰富的自然图像有益，但在形状主导的数据（如插画、草图）上会降低性能。现有方法虽尝试构建形状偏好模型，但缺乏可量化的指标来判断哪些数据集真正受益于此类改造。为此，本文提出一种数据驱动的度量方式：通过计算图像亮度（Y）通道与其L0平滑版本之间的结构相似性指数（SSIM），量化数据集中形状与纹理的相对占比。基于该度量，我们设计了一种计算高效的适应性方法，通过改变最大池化操作的膨胀率来增强形状偏好，同时保持卷积核权重冻结。实验表明，该方法在形状主导数据集上实现一致的性能提升，尤其在低数据条件下（仅需微调分类层）效果显著。
原文摘要 · Abstract (English)
Convolutional Neural Networks (CNNs) exhibit a well-known texture bias, prioritizing local patterns over global shapes - a tendency inherent to their convolutional architecture. While this bias is beneficial for texture-rich natural images, it often degrades performance on shape-dominant data such as illustrations and sketches. Although prior work has proposed shape-biased models to mitigate this issue, these approaches lack a quantitative metric for identifying which datasets would actually benefit from such modifications. To address this limitation, we propose a data-driven metric that quantifies the shape-texture balance within a dataset by computing the Structural Similarity Index (SSIM) between an image's luminance (Y) channel and its L0-smoothed counterpart. Building on this metric, we introduce a computationally efficient adaptation method that promotes shape bias by modifying the dilation of max-pooling operations while keeping convolutional weights frozen. Experimental results demonstrate consistent accuracy improvements on shape-dominant datasets, particularly in low-data regimes where full fine-tuning is impractical, requiring training only the final classification layer.
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。