人工智能培训

搜索

人工智能教程:RESPOND-CAM:通过可视化分析3D成像数据的深层模型(Respond-CAM: Analyzing Deep Models for 3D

[复制链接]
bigrc 发表于 2018-6-4 08:59:51 | 显示全部楼层 |阅读模式
bigrc 2018-6-4 08:59:51 501 0 显示全部楼层
人工智能教程:RESPOND-CAM:通过可视化分析3D成像数据的深层模型(Respond-CAM: Analyzing Deep Models for 3D Imaging Data by Visualizations)卷积神经网络(CNN)已成为各种生物医学图像分析任务的有力工具,但对CNN机器缺乏视觉解释。在本文中,我们提出了一种新颖的响应加权类激活映射(Respond-weighted Class Activation Mapping,Respond-CAM),用于通过可视化输入区域来解释基于CNN的模型,这些输入区域对预测非常重要,特别是对于生物医学三维成像数据输入。我们的方法使用流入卷积层的任何目标概念(例如目标类的分数)的梯度。加权特征映射被组合起来产生一个热图,该热图突出了图像中用于预测目标概念的重要区域。我们证明了Respond-CAM的一个较好的总和评分属性,并验证了它从当前最先进的方法对3D图像的显着改进。我们对细胞电子冷冻层析成像3D图像的测试表明,Respond-CAM在使用3D生物医学图像输入对CNN进行可视化方面实现了卓越的性能,并且能够在使用自然图像输入对CNN进行可视化方面获得相当好的结果。 Respond-CAM是一种高效可靠的CNN机械可视化方法,适用于各种CNN模型族和图像分析任务。
The convolutional neural network (CNN) has become a powerful tool for variousbiomedical image analysis tasks, but there is a lack of visual explanation forthe machinery of CNNs.In this paper, we present a novel algorithm,Respond-weighted Class Activation Mapping (Respond-CAM), for making CNN-basedmodels interpretable by visualizing input regions that are important forpredictions, especially for biomedical 3D imaging data inputs.Our method usesthe gradients of any target concept (e.g. the score of target class) that flowsinto a convolutional layer.The weighted feature maps are combined to produce aheatmap that highlights the important regions in the image for predicting thetarget concept.We prove a preferable sum-to-score property of the Respond-CAMand verify its significant improvement on 3D images from the currentstate-of-the-art approach.Our tests on Cellular Electron Cryo-Tomography 3Dimages show that Respond-CAM achieves superior performance on visualizing theCNNs with 3D biomedical images inputs, and is able to get reasonably goodresults on visualizing the CNNs with natural image inputs.The Respond-CAM isan efficient and reliable approach for visualizing the CNN machinery, and isapplicable to a wide variety of CNN model families and image analysis tasks.人工智能教程:RESPOND-CAM:通过可视化分析3D成像数据的深层模型(Respond-CAM: Analyzing Deep Models for 3D Imaging Data by Visualizations) J66AXb96aY20lL29.jpg
URL地址:https://arxiv.org/abs/1806.00102     ----pdf下载地址:http://arxiv.org/pdf/1806.00102    ----人工智能教程:RESPOND-CAM:通过可视化分析3D成像数据的深层模型(Respond-CAM: Analyzing Deep Models for 3D Imaging Data by Visualizations)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

bigrc当前离线
新手上路

查看:501 | 回复:0

快速回复 返回顶部 返回列表