RADAM：通过深度激活图的随机聚合编码进行纹理识别
RADAM: Texture Recognition through Randomized Aggregated Encoding of Deep Activation Maps
摘要：
纹理分析是计算机视觉中一个经典而又具有挑战性的任务，深度神经网络正在积极地应用于该任务。大多数方法都基于围绕预先训练的骨干构建特征聚合模块，然后根据特定的纹理识别任务微调新架构。在这里，我们提出了一种新的方法命名为\textbf{R} andom编码的\textbf{A}聚集\textbf{D} eep \textbf{A}激活\textbf{M} aps（RADAM），提取丰富的纹理表示，而不需要改变的骨干。该技术包括使用随机自动编码器（RAE）对预训练深度卷积网络的不同深度的输出进行编码。RAE使用封闭形式的解决方案在本地训练到每个图像，并且其解码器权重用于组成被馈送到线性SVM中的1维纹理表示。这意味着不需要微调或反向传播。我们在几个纹理基准上探索RADAM，并在不同的计算预算下实现了最先进的结果。我们的研究结果表明，预训练的骨干可能不需要额外的微调纹理识别，如果他们的学习表示更好地编码。
英文摘要：
Texture analysis is a classical yet challenging task in computer vision for which deep neural networks are actively being applied. Most approaches are based on building feature aggregation modules around a pre-trained backbone and then fine-tuning the new architecture on specific texture recognition tasks. Here we propose a new method named \textbf{R}andom encoding of \textbf{A}ggregated \textbf{D}eep \textbf{A}ctivation \textbf{M}aps (RADAM) which extracts rich texture representations without ever changing the backbone. The technique consists of encoding the output at different depths of a pre-trained deep convolutional network using a Randomized Autoencoder (RAE). The RAE is trained locally to each image using a closed-form solution, and its decoder weights are used to compose a 1-dimensional texture representation that is fed into a linear SVM. This means that no fine-tuning or backpropagation is needed. We explore RADAM on several texture benchmarks and achieve state-of-the-art results with different computational budgets. Our results suggest that pre-trained backbones may not require additional fine-tuning for texture recognition if their learned representations are better encoded.
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。
