Atrous 卷积，就是带洞的卷积，卷积核是稀疏的。
上图（b）是带洞卷积，可以跳着选，隔一个加一个。
下图中第三个示例（c），就是带洞卷积。
带洞卷积减少了核的大小，可以达到节省内存的作用。
而且带洞卷积的有效性基于一个假设：紧密相邻的像素几乎相同，全部纳入属于冗余，不如跳H(hole size)个取一个。
详细解决方案
相关解决方案
- 【解决】“Failed to get convolution algorithm. This is probably because cuDNN failed to initialize”
- 论文笔记--deeplabv3--Rethinking Atrous Convolution for Semantic Image Segmentation
- Diverse Neuron Type Selection for Convolution Neural Networks阅读笔记
- 论文 | Free-Form Image Inpainting with Gated Convolution
- C-COT:Beyond Correlation Filters: Learning Continuous Convolution Operators for Visual Trackin论文笔记
- Python-DNN-3D-Convolution(3D卷积)
- 转置卷积（Transposed Convolution）
- UnknownError: Failed to get convolution algorithm. This is probably because cuDNN failed to解决方案
- Understanding Convolution for Semantic Segmentation
- Separable Convolution和（regular）Convolution
- 【经典网络学习笔记2】AlexNet-ImageNet Classification with Deep Convolution Neural Network
- tensorflow.python.framework.errors_impl.UnknownError: Failed to get convolution algorithm. 解决方法
- （门控卷积实现）DeepFillv2（图像修复）：Free-Form Image Inpainting with Gated Convolution，pytroch代码实现
- Transposed convolution：pytorch ConvTranspose2d参数设置
- 2020-SIGIR- Lightgcn: Simplifying and powering graph convolution network for recommendation
- Correlation and Convolution
- 深入理解CVPR 2019论文 Linkage Based Face Clustering via Graph Convolution Network基于图卷积的人脸聚类
- CVPR 2019 Linkage Based Face Clustering via Graph Convolution Network论文翻译
- 【ESWIN实习】番外篇一、Convolution 2D kernel 官方文档编译流程（整理版）
- 【转载】Dilated Convolution 简单介绍
- 【转载】Atrous Convolution 简单介绍
- Convolution Layer卷积层理解
- Ufldl Exercise:Convolution and Pooling
- Failed to get convolution algorithm.This is probably because cuDNN failed to initialize
- 最容易理解的对卷积(convolution)的解释
- 迴旋濾波器(convolution filter)
- 扩张卷积（Atrous 卷积）
- DeepLab V3+——Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
- Pytorch：能够减少计算量的非对称卷积（asymmetric convolution）的pytorch实现
- Deconvolution layer或者Transposed convolution layer的理解