从拍摄的手写体数字串纸张照片到自动识别数字,本文从手写体数字分割、训练分类器和结果比较三方面结合过程图及代码解读手写体数字识别方法。
[数据挖掘] 回归和分类Kaggle实战
详细探讨解决Kaggle上某回归/分类比赛的全过程。包括用线性回归之梯度下降法/正规方程法(MATLAB)、神经网络之多层感知器(TensorFlow)、最近邻(scikit-learn)进行回归预测和用逻辑回归之梯度上升法(Python)、梯度提升决策树(XGBoost)进行分类预测。读者将理解线性回归和逻辑回归的原理/实现、其他框架的使用/调参,以及如何利用Python的多进程对逻辑回归的运算进行并行化提高效率。
[Web] Try Docker
Dive into the basics of Docker and learn how to create a self-contained, shareable development environment with containers, images, Dockerfiles, and volumes.
[Web] Spring Boot Web Application
Building a Spring Boot Web Application for movie ticket and service website.
[CVPR] Panorama Image Stitching
Automatically stiching several individual images to generate a panorama image.
Follow the routine described in the paper Automatic Panoramic Image Stitching using Invariant Features.
Theory and practice of panorama image stitching.
[Tensorflow] Human Pose estimation by Deep Learning
Summary and comprehension of papers: S.-E. Wei, V. Ramakrishna, T. Kanade, and Y. Sheikh. Convolutional pose machines. In CVPR, 2016.
A simple regression based implementation/VGG16 of pose estimation with tensorflow.
[CVPR] Image Morphing
Produce a “morph” animation of one image into another image, which involves two parts: cross dissolving and affine warping.
[Algorithm] Implement Split Function with KMP Algorithm
Splitting a string per the given seperator/delimiter similar to split() funtion in Python.
Implement with Knuth–Morris–Pratt string searching algorithm (or KMP algorithm) in C++ language.
[Caffe] Similarity-Based Deep Neural Network Hashing
A caffe version implementation of a hash network(DNNH/NINH) for similarity-based visual research based on paper: Hanjiang Lai, Yan Pan, Ye Liu, and Shuicheng Yan. Simultaneous feature learning and hash coding with deep neural networks, CVPR 2015.
[CVPR] A4 Paper Sheet Detection and Cropping with Hough Transform and Warping
Function similar to Document Scanner. Given images of A4 paper sheets, output paper sheets’ four corners as well as four edges and their equations. Then crop the background and leave the paper sheet in proper position and standard scaling. We can do this in three steps. Firstly, detect edges with hough transform. Then store the corners in order. Lastly, applying a perspective transform to warp the image.
[DIP] Histogram Equalization on Grayscale and Color Image
Histogram equalization is a technique for adjusting image intensities to enhance contrast. In this post, I implement grayscale image histogram equalization and three methods of color image histogram equalization. Detail analyses and results are given.
[PaperReading] Color Transfer between Images, Erik Reinhard et al, 2001
Implementation and improvement of paper: Color Transfer between Images, Erik Reinhard et al, 2001 —— A simple statistical analysis to impose one image’s color characteristics on another.
I put forward a method called PureColorGuidedStyle which is a easier, faster more controllable version based on this paper’s method. It can obtain better results most of the time.
[Assignment] Software Testing
Assignments of Software Testing course taught by Guoyang Cai, SYSU, 2017.
Assignment 1, 2, 3, 4 and 5
[Notes] Software System Analysis and Design
Notes of Software System Analysis and Design taught by Maolin Pan, SYSU, 2017.
Keypoints, assignments, courseware…
[Programming] Leetcode
Practice and summary.
001 TwoSum
002 AddTwoNumbers
003 Longest Substring Without Repeating Characters
[Notes] Guides by Andrej Karpathy
Some tips from two insightful guides by Andrej Karpathy.
[Notes] Programming Languages
Notes of Programming Languages, Part A by Dan Grossman, University of Washington in Coursera.
[Tips] About Blog
Some tips written down when I am blogging using Hexo, theme Next.
八皇后和八数码问题
Generate a large number of 8-puzzle and 8-queens instances and solve them(where possible) by hill climbing (steepest-ascent and first-choice variants), hill climbing with random restart, and simulated annealing. Measure the search cost and percentage of solved problems and graph these against the optimal solution cost. Comment on your results.
Hello World
What I cannot create, I do not understand