【语义分割】Fast-SCNN|【语义分割】Fast-SCNN -- Fast Semantic Segmentation Network
paper:Fast-SCNN
github:code
文章目录
- Abstract
- Fast-SCNN
- Network Architecture
- Pre-training on Auxiliary Tasks
- Experiment
- 1、Evaluation on Cityscapes
- 2、Pre-training and Weakly Labeled Data
- 3、Lower Input Resolution
Abstract 【【语义分割】Fast-SCNN|【语义分割】Fast-SCNN -- Fast Semantic Segmentation Network】Fast-SCNN是一个实时的语义分割模型。其基于现有的
two-branch
方法(BiSeNet),引入了一个learning to downsample
模块,在cityscapes上得到68.0%
的miou。FastSCNN采用
depthwise separable convolutions
和inverse residual blocks
Fast-SCNN 实时语义分割模型设计要点:
a larger receptive field
is important to learn complex correlations among object classes (i.e. global context)spatial detail
in images is necessary to preserve object
boundariesbalance speed and accuracy
![【语义分割】Fast-SCNN|【语义分割】Fast-SCNN -- Fast Semantic Segmentation Network](https://img.it610.com/image/info8/22af10bf1e014a2396254a5b7a97a9b6.jpg)
文章图片
![【语义分割】Fast-SCNN|【语义分割】Fast-SCNN -- Fast Semantic Segmentation Network](https://img.it610.com/image/info8/844053cbb51648b89722633abb40254d.jpg)
文章图片
PPM
是PSPNet中的一个重要模块,可融合 different-region-based context information.![【语义分割】Fast-SCNN|【语义分割】Fast-SCNN -- Fast Semantic Segmentation Network](https://img.it610.com/image/info8/b32175f2fbdb4502aaba55f8f642c960.jpg)
文章图片
Adding few layers after the feature fusion module boosts the accuracy
Pre-training on Auxiliary Tasks In our experiments we show that
small networks
do not get significant benefit from pre-training
. Instead, aggressive data augmentation and more number of epochs
provide similar resultsFast-SCNN在cityscapes上训练
1000
个epochExperiment 1、Evaluation on Cityscapes
![【语义分割】Fast-SCNN|【语义分割】Fast-SCNN -- Fast Semantic Segmentation Network](https://img.it610.com/image/info8/0f8ec1b716964a988bc3c1b85b828ab1.jpg)
文章图片
![【语义分割】Fast-SCNN|【语义分割】Fast-SCNN -- Fast Semantic Segmentation Network](https://img.it610.com/image/info8/e3b72c3b5a1d4b3280234afa4802800f.jpg)
文章图片
2、Pre-training and Weakly Labeled Data
![【语义分割】Fast-SCNN|【语义分割】Fast-SCNN -- Fast Semantic Segmentation Network](https://img.it610.com/image/info8/eaaa3a86f6f141678eabd035444dfbac.jpg)
文章图片
3、Lower Input Resolution
![【语义分割】Fast-SCNN|【语义分割】Fast-SCNN -- Fast Semantic Segmentation Network](https://img.it610.com/image/info8/70eb37160c53417197db20511fa0d145.jpg)
文章图片
目前很多模型输入都为512 x 1024
推荐阅读
- 宽容谁
- 我要做大厨
- 增长黑客的海盗法则
- 画画吗()
- 2019-02-13——今天谈梦想()
- 远去的风筝
- 三十年后的广场舞大爷
- 叙述作文
- 20190302|20190302 复盘翻盘
- 学无止境,人生还很长