原创 方向梯度直方圖(HOG)

1.介紹 HOG(Histogram of Oriented Gradient)是2005年CVPR會議上,法國國家計算機科學及自動控制研究所的Dalal等人提出的一種解決人體目標檢測的圖像描述子,該方法使用梯度方向直方圖(Histog

原创 batch size, mini-batch, iterations and epoch

Gradient descent is an iterative algorithm which computes the gradient of a function and uses it to update the parame

原创 論文閱讀筆記 -- ParsingNet

ParseNet: Looking Wider to See Better   The FCN approach can be thought of as sliding an classification network aroun

原创 Normalization vs Regularization in Machine or Deep learning

Normalisation adjusts the data; regularisation adjusts the prediction function. It is well-known that normalizing the i

原创 Logisitc Regrssion vs linear SVM

Given a binary classification problem, the goal is to find the “best” line that has the maximum probability of classify

原创 python multiprocessing 中imap和map的不同

本篇文章講python的multiprocessing中 imap、map、imap_unordered和map_async方法之間的區別。參考鏈接 主要有以下兩個區別: 它們使用你傳遞給它們的可迭代的對象的方式。 它們返回結果的方式。

原创 h5py vs npy/npz

Data engineering for computer vision Lately, I've been thinking hard about the best way to organize my data before feed

原创 map vs imap in multiprocessing.Pool

There are two key differences between imap/imap_unordered and map/map_async: The way they consume the iterable you pass

原创 Group Normalization vs Batch Normalization

BN 存在哪些問題: 1. BN 依賴大batch size, 當 batch size 太小時, batch statistics 變得不準確; 而顯存限制了batch size變大,尤其在檢測、分割等比較佔用顯存的模型上。 batch

原创 Histogram of Oriented Gradients

Histogram of Oriented Gradients DECEMBER 6, 2016 BY SATYA MALLICK In this post, we will learn the details of the

原创 Different types of SVM -- opencv

They are different formulations of SVM. At the heart of SVM is an mathematical optimization problem. This problem can

原创 Linear Kernel: Why is it recommended for text classification ?

The Support Vector Machine can be viewed as a kernel machine. As a result, you can change its behavior by using a diff

原创 argparse

1) argparse是python用於解析命令行參數和選項的標準模塊,用於代替已經過時的optparse模塊。 例如 python parseTest.py input.txt output.txt --user=name --po

原创 How a Kalman filter works, in pictures

I have to tell you about the Kalman filter, because what it does is pretty damn amazing. Surprisingly few softw

原创 Spatial Transformer Network

 STN架構中,分爲三個部分: Localisation netGrid generatorSampler Localisation net  把feature map U作爲輸入,過連續若干層計算(如卷積、FC等),迴歸出參數