arXiv 2602.072622026-02-10cs.CV

TwistNet-2D: 通过螺旋扭转学习第二阶通道交互用于纹理识别

TwistNet-2D: Learning Second-Order Channel Interactions via Spiral Twisting for Texture Recognition

McCormick School of Engineering, Northwestern University（西北大学工程学院）；School of Mathematics, University of Science and Technology of China（中国科学技术大学数学学院）；National University of Singapore（新加坡国立大学）；College of Artificial Intelligence, Nankai University（南开大学人工智能学院）；Guizhou University（贵州大学）；Information Initiative Center, Hokkaido University（北海道大学信息初始化中心）；School of New Energy, Jingchu University of Technology（荆楚科技大学新能源学院）；Wenzhou University（温州大学）

Junbo Jacob Lian, Feng Xiong, Yujun Sun, Kaichen Ouyang, Zong Ke, Mingyang Yu, Shengwei Fu, Zhong Rui, Zhang Yujun, Huiling Chen

备注：: Code is available at https://github.com/junbolian/TwistNet-2D

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

TwistNet-2D通过螺旋扭转学习第二阶通道交互，提升纹理识别性能，参数高效且超越多种基线模型。

摘要：

第二阶特征统计是纹理识别的核心，但当前方法面临根本性的矛盾：双线性池化和Gram矩阵捕捉全局通道相关性但会丢失空间结构，而自注意力模型则通过加权聚合来获取空间上下文，而不是显式地进行成对特征交互。我们引入TwistNet-2D，一个轻量级模块，该模块在方向性空间位移下计算局部成对通道乘积，共同编码特征何时共存以及如何相互作用。核心组件，螺旋扭转通道交互（STCI），在规定方向上位移一个特征图，然后进行元素级通道乘法，从而捕捉结构性和周期性纹理的跨位置共现模式。通过聚合四个方向头，利用学习的通道重加权，并将结果注入到具有sigmoid门控残差路径中，TwistNet仅在ResNet-18上增加3.5%的参数和2%的FLOPs，但始终在四个纹理和细粒度识别基准上超越了参数匹配和明显更大的基线模型——包括ConvNeXt、Swin Transformer和混合CNN-Transformer架构。

英文摘要：

Second-order feature statistics are central to texture recognition, yet current methods face a fundamental tension: bilinear pooling and Gram matrices capture global channel correlations but collapse spatial structure, while self-attention models spatial context through weighted aggregation rather than explicit pairwise feature interactions. We introduce TwistNet-2D, a lightweight module that computes \emph{local} pairwise channel products under directional spatial displacement, jointly encoding where features co-occur and how they interact. The core component, Spiral-Twisted Channel Interaction (STCI), shifts one feature map along a prescribed direction before element-wise channel multiplication, thereby capturing the cross-position co-occurrence patterns characteristic of structured and periodic textures. Aggregating four directional heads with learned channel reweighting and injecting the result through a sigmoid-gated residual path, \TwistNet incurs only 3.5% additional parameters and 2% additional FLOPs over ResNet-18, yet consistently surpasses both parameter-matched and substantially larger baselines -- including ConvNeXt, Swin Transformer, and hybrid CNN--Transformer architectures -- across four texture and fine-grained recognition benchmarks.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。