当前位置:网站首页>直观理解Dilated Convolution
直观理解Dilated Convolution
2022-07-20 05:32:00 【daimashiren】
Dilated Convolution是一种通过增加kernel元素间距(padding 0)使得感受野增加的一种卷积方式。所谓感受野就是特征图上每个点对应原图像的像素范围。
这样,在不改变kernel size(上图中的kernel size还是3x3)的情况下,增加了感受野。也可以理解为kernel size从3x3 变成了5x5(但是只有原来的3x3的位置有权值,其余位置均为0),Dilated Conv的kernel计算公式如下:
k ′ = k + ( k − 1 ) ( d − 1 ) k'=k+(k-1)(d-1) k′=k+(k−1)(d−1)
即在原来的k-1个间隔中充填d-1行(列)0得到一个新的k’。从而Dilated Conv得到的特征图的计算公式为:
H o u t = ⌊ H i n + 2 × padding − dilation × ( kernel_size − 1 ) − 1 stride + 1 ⌋ H_{out} = \left\lfloor\frac{H_{in} + 2 \times \text{padding} - \text{dilation} \times (\text{kernel\_size} - 1) - 1}{\text{stride}} + 1\right\rfloor Hout=⌊strideHin+2×padding−dilation×(kernel_size−1)−1+1⌋
Dilated Conv 在已有像素点的基础上,有意skip掉一些像素点,或者输入不变,但是Conv的kernel中某些权值部分设置为0,达到增大感受野的目的。
边栏推荐
- (1) Pytorch deep learning: linear model training
- Centos8(linux)安装mysql8.0.28
- (5) Pyqt5 series tutorials: use pychart to design the internal logic of pyqt5 in the serial port assistant parameter options (II)
- Redis publishing and subscription
- Centos8 (Linux) install mysql8.0.28
- 论文学习----URLLC benefit from NOMA (1)
- Differences between JDBC and ODBC
- SQL子查询
- [camp d'entraînement can] essai de mise en œuvre du Gan basé sur Shengsi
- Li Hongyi 2020 machine learning -- p17 CNN & P 14
猜你喜欢
Redis publishing and subscription
(1) Pytorch deep learning: linear model training
(4) Pyqt5 series tutorials: use pychart to design the internal logic of pyqt5 in the serial port assistant parameter options (I)
(六)PyTorch深度学习:Logistic回归(多层多维特征输入)
(二)PyTorch深度学习:梯度下降
MIMO-OFDM Wireless Communication Technology and matlab implementation (2) - Outdoor channel model under SISO
(4) Pytorch deep learning: pyttorch realizes linear regression
(9) Pytorch deep learning: convolutional neural network (the inception module in googlenet network architecture is this convolutional neural network architecture)
How to use idea to make jar packages
(一)PyTorch深度学习:线性模型训练
随机推荐
Paxos 论文 简单版 中文版
Swift archiving serialization
Thesis study ---- urllc benefit from noma (1)
昇腾工业质检应用实战
[deep learning] how to select the correct activation and loss function of the last layer for the model
Paper learning -- resource allocation in EE urllc in relay system
(10) Pytorch deep learning: convolutional neural network (simple residual convolutional neural network)
Endnote x9 import journal reference format of paper submission
【CANN訓練營】基於昇思的GAN實現隨筆
(九)PyTorch深度学习:卷积神经网络( GoogleNet网络架构中的 inception module 模块为本次卷积神经网络架构)
李宏毅2020机器学习--P11 Logistic Regression
论文阅读--Risk-Resistant Resource Allocation for eMBB and URLLC coexistence under M/G/1 Queueing Model
数仓基本概述
2D目标检测综述之检测模型篇(二)
Centos8 (Linux) 安装 redis
(一)PyTorch深度学习:线性模型训练
达梦收集统计信息
玩转CANN目标检测与识别一站式方案【介绍篇】
The role of 'defer' and 'async' attributes on the < srcipt> tag
论文学习--Resource allocation for multi-user downlink MISO OFDMA-URLLC systems