目标检测--Wide-Residual-Inception Networks for Real-time Object Detection
来源:互联网 发布:云杉网络 清华 李军 编辑:程序博客网 时间:2024/04/29 09:30
本文主要是在 Residual 和 Inception 基础上构建新的模型,实现实时物体检测。构建的新模型特色是占用内存少,计算量小。新网络命名为 wide-residual-inception (WR-Inception) 。
III. Wide-Residual-Inception Networks
A. Factors to Consider in Neural Network
Modeling—Width vs. Depth
文献【29】实验声称一个网络主要受三个因素影响:深度、宽度、滤波器尺寸。当时间复杂度固定,通过改变上面三个因素时,发现改变深度能最大的提升网络的性能。
文献【9】中的 wide-residual 宽残差网络证明通过残差网络中的 shortcut 链接,一个宽的浅层网络模型比 ResNets 效果要好。
本文主要针对嵌入式环境,结合前面两个文献的结论,提出了一个新模型,在一个大模型中重复使用一个小的模块。
B. Micro-Architecture
• Basic residual (3x3,3x3):
将两个3x3卷积层串联起来,再外加一个 shortcut
• Bottleneck (1x1, 3x3, 1x1):
首先用一个 1x1 卷积进行特征降维,再进行3x3卷积,最后再用 1x1恢复到原来的尺寸,外加一个 shortcut
• Inception:
该模块是 GoogleNet 中提出的,在同一个特征层上使用了不同种类的卷积,1x1, 3x3, 和 5x5,这么做可以提取不同尺度的目标特征。
C. Residual Inception Unit
上图 c 中显示的是 residual-inception 模块,它对 inception 模块加了一个 shortcut,首先进行 1x1 卷积,再分叉进行 3x3, 和 5x5 卷积,而不是对每个路线进行 1x1 卷积。这里 5x5 卷积 是用两个3x3卷积来表示的。最后使用 concatenation 将这些组合起来。效果如下图所示:
D. Macro-Architecture
上面三个网络的计算量大致一样,但是网络性能不一样。
这里使用 a Single-Shot Multi-box Detector (SSD) 这个框架来论证上面模型的效果。可以检测不同尺寸物体。
计算量对比图:
效果:
- 目标检测--Wide-Residual-Inception Networks for Real-time Object Detection
- 目标检测--PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- 目标检测--PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- 目标检测--Feature Pyramid Networks for Object Detection
- 目标检测“Feature Pyramid Networks for Object Detection”
- 目标检测“Perceptual Generative Adversarial Networks for Small Object Detection”
- 目标检测“Perceptual Generative Adversarial Networks for Small Object Detection”
- 目标检测“Perceptual Generative Adversarial Networks for Small Object Detection”
- 嵌入式目标检测--Fast YOLO: A Fast You Only Look Once System for Real-time Embedded Object Detection
- READING NOTE: PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection - arxiv 2016.08
- PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- 论文笔记:PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- [Paper note] PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- 读书笔记PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- oracle中varchar,varchar2,nvarchar,nvarchar2之间的区别
- Unknown class ViewController in Interface Builder file.
- ubuntu+cuda+theano
- was not granted this permission: android.permission.WRITE_SETTINGS.
- Android Activity使用详解
- 目标检测--Wide-Residual-Inception Networks for Real-time Object Detection
- 华为机试 取近似值
- postman使用教程
- Gson的使用(2)
- Fiddler (三) Composer创建和发送HTTP Request
- springboot打包docker镜像部署
- QFile和QDir使用以及自动文件创建
- TextView显示不同颜色文字--工具类TextUtil
- AI学习之路(1):TensorFlow在Windows安装