Download Data training set (138G)
validation set (6.3G-50000 sheets)
train_label.txt
validation_label.txt
p.s with thunderbolt is quite fast, 3 days
Data Decompression tar xvf ILSVRC2012_img_train.tar -C ./train
tar xvf ILSVRC2012_img_val.tar -C ./val
For train dataset, after decompression, there are 1000 tar files, which need to be decompressed again. The decompression script unzip.sh is as follows
dir=/data/srd/data/Image/ImageNet/train for x in `ls $dir/*tar` do filename=`basename $x .tar` mkdir $dir/$filename tar -xvf $x -C $dir/$filename done rm *.

This article records Fibonacci number's Python implementation: recursion and loop solutions, as well as some useful topics. Python Implementation recursion according to the traditional recursive way, concise and elegant.But it is written out as an algorithm of o (N2) o (n 2) o (N2) def fibo(n): """肥波那契函数""" if n < 3: return 1 else: return fibo(n-1) + fibo(n-2) O(n)O(n)O(n)

work needs, studied some very classic image retrieval algorithms, recorded one by one, to facilitate their review and communication.
This blog post is about VLAD (Vector of Locally Aggregated Descriptors), that is, the vector of aggregated local descriptors. It is a process of using local descriptors of images such as SIFT, SURF, ORB, etc. to do some aggregation operations, and then using a long vector to represent an image.Representation of images into vectors is a prerequisite for image retrieval.

topic descriptionOn this topic, the premise is that n steps will have an n-order jump method.The analysis is as follows: f (1) = 1f (2) = f (2-1)+f (2-2)//f (2-2) represents the number of times that the 2nd order jumps the 2nd order at a time. f(3) = f(3-1) + f(3-2) + f(3-3) ... f(n) = f(n-1) + f(n-2) + f(n-3) + ... + f(n-(n-1)) + f(n-n) Description: 1）

I believe there will be some doubts about the search problem. This is the best time and the worst time. We missed bitcoin in 13 years, Ethereum in 14 years, Ripple in 15 years, ant coins and so on. We missed many opportunities to become rich. We didn't have the chance to become rich second generation, but we had the chance to make our children rich second generation.
This question needs a serious answer.

Some students reported that each article was written too short and read too fast.On these two issues, after I was in Thought for a long time, I also felt that it was necessary to say something.Yes-from Lu Xun's "Commemoration of the Prince of Liu Hezhen".As a man, so many people use short and fast to describe me, you can understand meThis kind of sad?When this reputation comes out, who will be willing to do with me and love to do things ~ ~ ~ right, I also wish I couldEnough to show you, prove that I am a real man, a L'homme du printemps.

niuke net exercise portal topic descriptionSolutionMethod 1: Ordinary recursion, but Call stack overflow is easy to occur, and the speed will be slow as the number of call layers increases.The most obvious thing is that Niuke's recursive version on the Internet does not work. function Fibonacci(n) { /* 方法一 递归，但是容易出现调用栈溢出的情况*/

Is this really the Map I know? Open the door and ask a few questions first, so the purposeful research will be clearer.1. Have you used any other Map besides HashMap?2. How is the bottom layer of HashMap implemented?Is map disorder?4. How does Concurrent HashMap achieve thread safety? Have you used any other Map besides HashMap? interesting HashMap is very common in practice, even we often equate HashMap with Map.What

Problem Description: A step has N steps. If you can jump 1 step at a time, you can also jump 2 steps. Find out how many jump methods there are in total
int ClimbStairs(int n) { int dp[3]={1,1}; if(n<2) { return 1; } for(int i=2;i<=n;i++) { dp[2]=dp[0]+dp[1]; dp[0]=dp[1]; dp[1]=dp[2]; } return dp[2]; }