Context-aware padding semantic segmentation ----tc

最新推荐文章于 2024-08-07 14:52:30 发布

tcheng1234

最新推荐文章于 2024-08-07 14:52:30 发布

阅读量167

点赞数 2

分类专栏：深度学习文章标签： html5

本文链接：https://blog.csdn.net/tcheng1234/article/details/120462250

版权

深度学习专栏收录该内容

1 篇文章 0 订阅

订阅专栏

Research motivation :

For semantic segmentation we use zero padding to prevent feature map will diminish fast,but the and also lose the border information.

The insufficient Zero padding : for the learned filter(include learn from border will share in all spatial locations)

So ,they want to use context-aware (CA) to padding:

To know what the first layer receive,only taking local region to directly predict the displacement to extrapolate the image. (the image extrapolation)

The main contributions:

new padding method according to image extrapolation.
Compare with sota method in SS task.

Method

Network:

Algorithm:

Assume the $D_{input}$ as the input image

$For : X_i = randomcope(D_{input) \\$

$:\quad b = mask(X_i)$

$(x_i,y_i) = pixel\,location(b)\\ (\Delta x,\Delta y) = D(E(b,W))\\ x_i^{warp}, y_i^{warp} = x_i+\Delta x,y_i+\Delta y\\ b_i^{i} = \sum_{k\epsilon N(x_i,y_i))}b_k(1-\left \| x_i-x_k \right \|)(1-\left \| y_i-y_k \right \|) \quad N(x_i,y_i) is \, 4 neighborhoods \, of (x_i,y_i)\\ Loss:\\ L = \frac{1}{N}\sum_{b\varepsilon D}(\left \| b\bigodot (1-M)-T(b\bigodot M,F)\bigodot (1-M) \right \|) \quad where\, D \,is\, the\, training \,set, N\, is \,its\, cardinality\\ backpropagation \\ end.\\ hnit:padding\, model T,$

$(\Delta x,\Delta y) \epsilon F\\ F\, is\, the\, displacement\, predicted by\, E\\ E \,is\, ecoder-decoder\, architecture\\$

Expermients：

Datasets	Cityscape	DeepGlobe satellites
total	5000	1146
train_set	2975	803
test_set	1525	172
vaildation set	500	171
numer_calss	19	7

Metric: mlou

Segmeantation network :PSPNet

Conclusion :

higher mlou than zero padding.

The point:

pyramid pooling : Multi-scale features are extracted from fixed size feature vectors.

Know the difference of SPP-Net and R-CNN

atrous spatial pooling: first be put forward in Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs in the condition of keeping feature map sable to increase receptive-field

Image regeneration :

GANs based model
autoregressive models

The padding way:

zero padding
Reflection padding
circular padding ---- >re-weighting based scheme called prartial convolution
mean distribution padding
symmetric padding

The inspiration:

to care more border information can lead to better segmentation result.the moudle maybe can be used !!!!

tcheng1234

关注

2
点赞
踩
0

收藏

觉得还不错? 一键收藏
0
评论
Context-aware padding semantic segmentation ----tc

Research motivation :For semantic segmentation we use zero paddingto prevent feature map will diminish fast,but the and also lose the border information.The insufficient Zero padding: for the learned filter(include learn from border will share in all..
复制链接

扫一扫