The Star Also Rises

Monday, November 25, 2019

FC-CRF

FC-CRF

2019/11/18

前言：

傳統的圖像分割和標記技術是在像素或者圖片區域上定義 CRF。雖然區域模型上通常有密集的成對連通性，但是像素級別的模型一般只允許用在稀疏圖結構，CRF 圖模型的邊非常多, 不能使用傳統方法，但用近似算法推斷全連接的 FC-CRF 模型，其中成對邊緣的勢能由高斯核的線性組合定義。

-----

# FC-CRF

-----

References

# FC-CRF

Krähenbühl, Philipp, and Vladlen Koltun. "Efficient inference in fully connected crfs with gaussian edge potentials." Advances in neural information processing systems. 2011.

http://papers.nips.cc/paper/4296-efficient-inference-in-fully-connected-crfs-with-gaussian-edge-potentials.pdf

语义分割中 CRF 的运用 _ Mooyu's Blog
http://guoxs.github.io/Blog/2018/03/10/CRF/

FCN(3)——DenseCRF - 知乎
https://zhuanlan.zhihu.com/p/2246458

DRN

DRN

2019/11/18

與 ENet 同樣使用 ResNet 與擴張卷積的 DRN，改善擴張卷積的缺點，論文作者稱為 degridding。三個方法分別是：移除 max pooling、增加殘差層數、移除 7、8 兩層的恆等映射。

-----

# DRN

-----

# DRN

-----

References

# DRN

Yu, Fisher, Vladlen Koltun, and Thomas Funkhouser. "Dilated residual networks." Proceedings of the IEEE conference on computer vision and pattern recognition. 2017.

http://openaccess.thecvf.com/content_cvpr_2017/papers/Yu_Dilated_Residual_Networks_CVPR_2017_paper.pdf

空洞卷积（Dilated Convolution）：有之以为利，无之以为用 - 知乎
https://zhuanlan.zhihu.com/p/66796313

ENet

ENet

2019/11/18

前言：

以 SegNet 為基線，ENet 在 ResNet 的架構上整合了傳統卷積、Inception v3 的非對稱卷積、與 DilatedNet 的擴張卷積，在速度上取得極佳的表現，但在準確率上則不佳。可說是 Semantic Segmentation 上的 YOLO。

-----

# ENet

-----

# DilatedNet

-----

// AlphaTree-graphic-deep-neural-network_InceptionV3.md at master · weslynn_AlphaTree-graphic-deep-neural-network · GitHub

-----

# ENet

-----

References

# ENet

Paszke, Adam, et al. "Enet: A deep neural network architecture for real-time semantic segmentation." arXiv preprint arXiv:1606.02147 (2016).

https://arxiv.org/pdf/1606.02147.pdf

# DilatedNet

Yu, Fisher, and Vladlen Koltun. "Multi-scale context aggregation by dilated convolutions." arXiv preprint arXiv:1511.07122 (2015).

https://arxiv.org/pdf/1511.07122.pdf

-----

AlphaTree-graphic-deep-neural-network_InceptionV3.md at master · weslynn_AlphaTree-graphic-deep-neural-network · GitHub
https://github.com/weslynn/AlphaTree-graphic-deep-neural-network/blob/master/object%20classification%20%E7%89%A9%E4%BD%93%E5%88%86%E7%B1%BB/InceptionV3.md

-----

ENet —一种针对实时语义分割的深度神经架构 - 知乎
https://zhuanlan.zhihu.com/p/57720412

DilatedNet

DilatedNet

2019/11/18

前言：

與傳統卷積相比，擴張卷積的計算量相同，然而可以取得較大的視野。

-----

# DilatedNet

-----

References

# DilatedNet

Yu, Fisher, and Vladlen Koltun. "Multi-scale context aggregation by dilated convolutions." arXiv preprint arXiv:1511.07122 (2015).

https://arxiv.org/pdf/1511.07122.pdf

DilatedNet - 扩张卷积（语义分割） - 知乎
https://zhuanlan.zhihu.com/p/61304801

SegNet

SegNet

2019/11/18

前言：

SegNet 的編碼器使用 VGG16。與 DeconvNet 相比，SegNet 進行更多性能、記憶體、速度上的平衡。首先 SegNet 去掉了 FC 層，所以參數量降低為 1/10，另外在 Upsampling 的時候，使用 Conv 層而非 Deconv 層。

-----

# SegNet

-----

# SegNet

-----

References

# SegNet

Badrinarayanan, Vijay, Alex Kendall, and Roberto Cipolla. "Segnet: A deep convolutional encoder-decoder architecture for image segmentation." IEEE transactions on pattern analysis and machine intelligence 39.12 (2017): 2481-2495.

https://arxiv.org/pdf/1511.00561.pdf

SegNet论文笔记 - 知乎
https://zhuanlan.zhihu.com/p/36525939

DeconvNet

DeconvNet

2019/11/18

前言：

DeconvNet 是對稱的結構，由於並未取消全連接層，所以參數量還是很大。解碼器部分主要由 unpooling 和 deconvolution 構成，unpooling 通過記錄解碼器對應的 max pooling 的 indice 取得原始結構，其結果是稀疏的，使用 deconvolution 可進行學習。

-----

# DeconvNet

-----

# DeconvNet

-----

References

# DeconvNet

Noh, Hyeonwoo, Seunghoon Hong, and Bohyung Han. "Learning deconvolution network for semantic segmentation." Proceedings of the IEEE international conference on computer vision. 2015.

https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Noh_Learning_Deconvolution_Network_ICCV_2015_paper.pdf

DeconvNet & SegNet
https://jimchenhub.github.io/posts/2019-04-16-DeconvNet%20&%20SegNet

FCN

FCN

2019/11/18

前言：

FCN 是語意分割（Semantic Segmentation）的經典論文。舉例說明，照片上，屬於狗的像素，都會被塗上同一顏色，貓的像素則被塗上另一顏色，其餘類推。傳統上有條件隨機場（CRF）等方法，FCN 則把卷積神經網路最後的全連接層也換成卷積層，最後的特徵圖因而保留概略的輪廓，然後上取樣兩次，達到不錯的效果。值得注意的是，使用 VGG16 比 GoogLeNet 的效果好不少。

-----

# FCN

-----

// Fully Convolutional Networks _ 简说

-----

References

# FCN

Long, Jonathan, Evan Shelhamer, and Trevor Darrell. "Fully convolutional networks for semantic segmentation." Proceedings of the IEEE conference on computer vision and pattern recognition. 2015.

https://www.cv-foundation.org/openaccess/content_cvpr_2015/papers/Long_Fully_Convolutional_Networks_2015_CVPR_paper.pdf

FCN学习:Semantic Segmentation - 知乎
https://zhuanlan.zhihu.com/p/22976342

Fully Convolutional Networks _ 简说
http://simtalk.cn/2016/11/01/Fully-Convolutional-Networks/

Monday, November 18, 2019

nms()

nms()

2019/10/07

nms(): perform non-maximal suppression on the results above along with appropriate threshold for corner detection

-----

References

structure_tensor()

structure_tensor()

2019/10/07

structure_tensor(): use the gradient magnitude above to compute the structure tensor (second-moment matrix).

References

Structure tensor - Wikipedia
https://en.wikipedia.org/wiki/Structure_tensor

Structure Tensor - Tutorial and Demonstration of the uses of structure tensors in gradient representation
https://www.cs.cmu.edu/~sarsen/structureTensorTutorial/

图像结构张量(structure tensor) - OO - CSDN博客
https://blog.csdn.net/Alphabetic/article/details/80329480

sobel_edge_detection()

sobel_edge_detection()

2019/10/07

sobel_edge_detection(): apply the Sobel filters to the blurred images and compute

-----

// How Image Edge Detection Works - Aryaman Sharda - Medium

-----

// Edge detection with the Sobel operator in Ruby – saush

-----

// First-order Derivative kernels for Edge Detection _ TheAILearner

-----

// Sobel edge detector with scikit-image - Hands-On Image Processing with Python [Book]

-----

// Edge detection by Sobel operator - Rajeev Ranjan - Medium

-----

Step edge
Ramp edge
Ridge edge
Roof edge

// OpenCV – more on edge detection – CH.Tseng

-----

Laplacian
Sobel
Canny
Prewitt
Scharr

-----

-----

References

◎ Paper

Sobel, Irwin, and G. Feldman. "An Isotropic 3x3 Image Gradient Operator for Image Processing." Mach. Vis. Three–Dimens. Scenes June (1968): 376-379.

https://www.researchgate.net/publication/239398674_An_Isotropic_3x3_Image_Gradient_Operator/link/557e06f508aeea18b777c389/download

-----

Sobel operator - Wikipedia
https://en.wikipedia.org/wiki/Sobel_operator

# 綜述
# 107 claps
What is Edge Detection Convolutional Neural Network’s (CNN’s) edge detection
https://medium.com/@ayeshmanthaperera/edge-detection-7e09bb3531d9

# 97 claps
Understanding Edge Detection (Sobel Operator) - Data Driven Investor - Medium
https://medium.com/datadriveninvestor/understanding-edge-detection-sobel-operator-2aada303b900

How Image Edge Detection Works - Aryaman Sharda - Medium
https://medium.com/@aryamansharda/how-image-edge-detection-works-b759baac01e2

# 52 claps
Canny Edge Detection Method - Aydin Ayanzadeh - Medium
https://medium.com/@a.ayanzadeh/canny-edge-detection-method-23a23b282ac0

# 299 claps
Canny Edge Detection Step by Step in Python — Computer Vision
https://towardsdatascience.com/canny-edge-detection-step-by-step-in-python-computer-vision-b49c3a2d8123

Edge detection with the Sobel operator in Ruby – saush
https://blog.saush.com/2011/04/20/edge-detection-with-the-sobel-operator-in-ruby/

Sobel edge detector with scikit-image - Hands-On Image Processing with Python [Book]
https://www.oreilly.com/library/view/hands-on-image-processing/9781789343731/0d1fdc9a-8833-43d2-9e28-f89185793e7f.xhtml

// First-order Derivative kernels for Edge Detection _ TheAILearner
https://theailearner.com/2019/05/24/first-order-derivative-kernels-for-edge-detection/

-----

討論索伯算子 Sobel 的 Convolution 問題 - Rifur Ni - Medium
https://medium.com/@Rifur/%E8%A8%8E%E8%AB%96%E7%B4%A2%E4%BC%AF%E7%AE%97%E5%AD%90-sobel-%E7%9A%84-convolution-%E5%95%8F%E9%A1%8C-e986ce026e72

OPENCV – Edge detection邊緣偵測 – CH.Tseng
https://chtseng.wordpress.com/2016/12/05/opencv-edge-detection%E9%82%8A%E7%B7%A3%E5%81%B5%E6%B8%AC/

OpenCV – more on edge detection – CH.Tseng
https://chtseng.wordpress.com/2016/12/05/opencv-more-on-edge-detection/

邊界偵測 (Edge Detection) – Ya-Ti's Blog
https://blog.ya-ti.nctu.me/2018/07/edge-detection/?i=1

Edge Detection Algorithm using OpenCV.md · GitHub
https://gist.github.com/MarcWang/52bd9e0dea9e168a010c7e83d7f2b3b0

You are entering my brain...... Edge Detection--Sobel Filter
http://tkdstkdstkds.blogspot.com/2012/02/edge-detection-sobel-filter.html

Homework Show @ IPC HW7 - 邊緣偵測 (Sobel operator)。
http://hwshow-ipc.blogspot.com/2010/01/hw7-sobel-operator.html

CS362_1001545 [HW3] Sobel Edge Detection
http://yzu1022cs362s1001545.blogspot.com/2014/03/hw3-sobel-edge-detection.html

Edge detection by Sobel operator - Rajeev Ranjan - Medium
https://medium.com/@rajeevranjan_47004/edge-detection-by-sobel-operator-ba9c90b001c0

-----

代碼實作

# 312 claps
Comparing Edge Detection Methods - Nika Tsankashvili - Medium
https://medium.com/@nikatsanka/comparing-edge-detection-methods-638a2919476e

Sobel edge detector _ Martin
https://cmmlin2016.wordpress.com/2013/11/18/sobel-edge-detector/

gaussian_smooth()

gaussian_smooth()

2019/10/07

-----

gaussian_smooth(): filter images with Gaussian blur.

-----

Gaussian blur is a filter used to smooth or blur a digital picture. It replaces pixels with a weighted average of surrounding pixels. The weights come from the Gaussian probability distribution, so the nearest pixels are more influential.

-----